Development of vibration-based RV reducer diagnostic criteria and factory evaluation standards Authors: Xin Wang, Jiangwen Wei, Yongliang Ding, Xiaobin Liu, Feng Zhang Published: 2026 ABSTRACT To address the current absence of vibration characteristic factory inspection standards for RV reducer manufacturers, this study concentrates on the RV40E reducer. By integrating factory inspection procedures and test bench systems with large-scale vibration signal analysis from production lines, we established a time-frequency domain evaluation standard for vibration-based product qualification through extensive sample analysis. The standard addresses 90% of typical factory defects, encompassing eccentricity errors, gear imperfections, and inspection inaccuracies. This approach allows for precise identification of specific fault types in non-conforming products using vibration analysis, thereby aiding in product repairs and performance optimization. The research has been successfully validated in small-scale enterprise production. Experimental results demonstrate that the proposed evaluation standard achieves 96.7% accuracy in fault identification under complex operating conditions, significantly outperforming traditional methods. This approach offers substantial practical value by resolving the industry’s lack of standardized vibration testing protocols. With all samples sourced from actual production lines, the data demonstrates high reliability and practical applicability, offering valuable references for the development of the RV reducer industry.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-55040-8. Introduction As the core component for industrial robot joint drives, RV reducers dictate the overall performance of robotic systems through their transmission accuracy, reliability, and longevity. With the growing demands of intelligent manufacturing on the comprehensive performance of industrial robots, the requirements for RV reducer performance have become increasingly stringent. Traditional manual inspection methods can no longer meet the factory-level consistency testing needs. Thus, the introduction of vibration detection online monitoring systems, which utilize vibration signals to provide more precise performance evaluation metrics, will significantly enhance product quality and enable continuous improvement in enterprise manufacturing. In recent years, research on fault diagnosis and factory certification of RV reducers has gained momentum. Early studies primarily focused on issues such as improper tooth profile shaping1, dynamic model errors2, and torsional stiffness3, aiming to refine the design of RV reducers. While these studies provided support for optimizing reducer design, a comprehensive fault diagnosis system tailored for engineering applications has yet to be established. With the advancement of industrial robot industrialization, the demand for engineering transformation and early design optimization research has driven a shift in research focus. Currently, priorities have shifted towards practical operational challenges, including product errors, high-frequency failures, and core component failures. The internal structure of the RV reducer is complex, encompassing multiple precision meshing pairs such as cycloidal pin gears4,5, planetary gears, eccentric axes, and critical bearings6–8. Due to prolonged operation under high loads, high speeds, frequent start-stop cycles, and varying working conditions, RV reducers are susceptible to various mechanical failures, including wear, pitting, and tooth breakage9. In the research on dynamic behavior monitoring and damage diagnosis for RV reducers, transmission error stands as an indispensable core quantification indicator for calibrating motion accuracy and assessing geometric defects10,11. However, the detection of transmission errors depends on specialized fixtures and labor-intensive methods, rendering it unsuitable for inspections on mass production lines. In actual manufacturing, the manual judgment of periodic abnormal noises continues to be a common practice, despite its high subjectivity, low efficiency, and frequent missed detections. Particularly in batch production, maintaining consistent inspection standards proves challenging. Leveraging the significant advantages of vibration signals in practical detection, vibration characteristic analysis methods combined with unified inspection standards derived from batch production test data have emerged as a critical solution to these issues. Vibration signals12,13 can reflect nonlinear behaviors, such as fluctuations in meshing stiffness and variations in bearing clearance, in RV reducers under variable loads in real-time. This feature makes them an ideal medium for fault diagnosis in complex working conditions. It has spurred domestic and international scholars to develop core technical systems, including time-frequency analysis and machine learning models, which offer direct methodological support for fault feature extraction and diagnostic model construction in RV reducers. However, the unique dynamic characteristics of RV reducers, such as rigid coupling and significant nonlinear vibrations, present adaptation challenges to these general technical systems. This motivates researchers to develop enhancements and innovative applications tailored to their specific characteristics. To address the unique dynamic characteristics of RV reducers, researchers have implemented adaptive improvements to the aforementioned technical framework, innovatively applying these enhancements to fault diagnosis. In dynamic modeling, Yi et al.14 aimed to reveal the vibration characteristics of RV reducers under partial failure conditions of cycloidal and needle gears. By considering the internal contact relationships within RV reducers, they established a dynamic model to study vibrations at fixed housings. For cycloidal-needle gear stages, a specialized method was proposed to calculate the time-varying meshing stiffness under healthy and faulty conditions. Incorporating the impact of partial failures in cycloidal and needle gears on meshing stiffness into the RV reducer’s dynamic model, they simulated fault-induced vibrations. Zheng et al.15 developed a lumped-parameter model for RV reducers to address data scarcity issues, effectively overcoming the problem of insufficient fault samples. In assembly precision control, Du et al.16 proposed a cycloidal tooth profile compensation method to mitigate machining errors. Chu et al.17 applied a genetic algorithm-based approach to selective assembly in RV reducers, ensuring assembly accuracy. In fault diagnosis, Yang et al.18 utilized acoustic emission signals, which are more sensitive to minor faults, for fault diagnosis in RV reducers. Qiao et al.19 proposed a nonlinear spectral feature fusion method for fault diagnosis in industrial robot RV reducers. While the aforementioned studies have enhanced theoretical analytical precision, they operate under idealized working conditions, resulting in a modal mismatch with the actual production line environment characterized by multi-noise and variable loads. Although research predominantly focuses on theoretical modeling and static optimization of RV reducer transmission accuracy, there remains a significant gap in translating these theoretical achievements into practical fault diagnosis applications. As research perspectives shift from single-precision improvements to comprehensive fault diagnosis, the limitations of existing technical frameworks become increasingly apparent, urgently requiring systematic solutions to bridge this critical gap. This study concentrates on the engineering application of vibration characteristics in RV reducers. By combining theoretical analysis with experimental testing, we have developed an artificial discrimination method that utilizes the time-frequency features of vibration signals, offering objective evidence to support manual detection. Through large-sample statistical analysis and correlation studies, we have proposed a factory quality evaluation standard system based on vibration characteristics. The research has been implemented in small-batch production line testing to facilitate human-assisted diagnosis. In the future, we plan to refine these standards to support large-scale production line testing. By developing a theoretical index generation system software and integrating a residual network20, a particle swarm optimization algorithm based on EEMD21, and TPA/EMD envelope analysis techniques22, our goal is to further develop an intelligent automated detection system. The foundational data test results from this research provide industry-wide reusable benchmark samples, offering significant reference value for advancing the practical application of related technologies in engineering. Experimental testing equipment All data in this study were sourced from product inspection samples collected at corporate factories. The fault cases analyzed were derived from real-world production errors, with no fabricated samples involved. The research focused on testing and analyzing the RV40E model, the company’s top-selling product. By comparing vibration test results with the enterprise’s factory inspection parameters (including noise, backlash, and transmission error), we established reasonable vibration signal standards for factory testing. Test bench This study conducts experimental testing and research on the RV40E series of reducers. The series offers four gear ratios: 81, 105, 121, and 153, with the most common being 81 and 121. To validate the comprehensiveness and universality of theoretical research, this paper performs tests and comparative analyses on these two ratios. The experimental testing for this study was conducted in the factory’s inspection laboratory, with the test bench shown in Fig. 1. The RV reducer was mounted on a fixed bracket to connect the motor and the RV reducer. The authors conducted multiple tests to determine optimal installation positions for the accelerometer, including the cylindrical surface of the gearbox, output-end fixing ring, rotating surfaces, and robotic arm positions. Comparative tests across various locations revealed that installing the accelerometer directly above the fixed bracket yielded the best test results, as illustrated in Fig. 1a. This position mimics a cantilever beam configuration, providing signal amplification with clear signals in all three directions and minimal interference. Other installation positions exhibited limitations such as unstable signal stability, significant directional variations, and weak signal strength in certain directions. The sensor coordinate system is depicted in Fig. 1, where X represents the horizontal direction, Y the vertical direction, and Z the axial direction. The three-axis accelerometer demonstrated a sensitivity of 100mV/g, measurement range of 1–10 kHz, and sampling frequency of 5000 Hz, with magnetic mounting method. Fig. 1RV reducer factory test bench. (a) Directly above the fixed bracket; (b) Fixed ring at the output end; (c) Cylindrical surface of the gearbox, and (d) Axial view of the robotic arm shaft. RV reducer factory test bench. (a) Directly above the fixed bracket; (b) Fixed ring at the output end; (c) Cylindrical surface of the gearbox, and (d) Axial view of the robotic arm shaft. Reducer parameters The reducer structure examined in this study is depicted in Fig. 2. The central axis and planetary gear constitute the first stage of gear meshing, while the gear pin and cycloidal gear form the second stage. The motor, which is connected to the central axis, acts as the input end. To ensure the universality of the proposed detection method, two primary gear ratios of the RV40E models were chosen for investigation. These gear ratios vary in tooth count, with specific parameters outlined in Table 1. The standardized description of the reducer adheres to the format RV40E-121, where RV40E represents the model and 121 signifies the gear ratio. The test was carried out with a motor input speed of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${v_1}$$\end{document}=2000 rpm, rotating forward for 5–10 s, then reversing, and repeating this cycle three times. Considering that gear ratios determine the number of teeth across gear stages, they produce unique characteristic frequencies for each ratio, as detailed in Table 2.This frequency data will serve as the basis for diagnosing issues related to the reducer, such as vibration, noise, and abnormal sounds, in subsequent analysis. Fig. 2Schematic diagram of RV reducer structure. Schematic diagram of RV reducer structure. Table 1Gear ratios for RV40E at various speed ratios.Number of teethGear ratio r81121Center axis gear teeth z11612Planetary gear teeth z23236Cycloidal gear teeth z33939Gear pin number z44040 Table 1Gear ratios for RV40E at various speed ratios.Number of teethGear ratio r81121Center axis gear teeth z11612Planetary gear teeth z23236Cycloidal gear teeth z33939Gear pin number z44040 Gear ratios for RV40E at various speed ratios. Table 2Characteristic frequencies of gears at all stages.Characteristic frequencyFormular = 81r = 121Central axis rotation frequency f1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_1}=\frac{{{v_1}}}{{60}}$$\end{document} 33.33 Hz33.33 HzPlanetary gear rotation frequency f2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_2}=\frac{{{z_1}{z_4}}}{{\left( {{z_3} - {z_4}} \right)\left( {{z_1}+{z_2}{z_4}} \right)}}{f_1}$$\end{document} −16.46 Hz−11.02 HzPlanetary gear rotation period T2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T_2}=\frac{1}{{{f_2}}}$$\end{document} 0.06s0.09sCycloid gear rotation frequency f3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_3}=\frac{{{{\mathrm{z}}_1}}}{{{{\mathrm{z}}_1}+{{\mathrm{z}}_2}{{\mathrm{z}}_4}}}{f_1}$$\end{document} 0.41 Hz0.28 HzCycloidal gear rotation period T3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T_3}=\frac{1}{{{f_3}}}$$\end{document} 2.44s3.57sSecond stage meshing frequency f2n \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_{2{\mathrm{n}}}}=\frac{{{{\mathrm{z}}_1}{{\mathrm{z}}_3}{{\mathrm{z}}_4}}}{{{{\mathrm{z}}_1}+{{\mathrm{z}}_2}{{\mathrm{z}}_4}}}{f_1}$$\end{document} 641 Hz429 HzFirst stage meshing frequency f1n \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_{1{\mathrm{n}}}}=\frac{{{{\mathrm{z}}_1}{{\mathrm{z}}_2}{{\mathrm{z}}_4}}}{{{{\mathrm{z}}_1}+{{\mathrm{z}}_2}{{\mathrm{z}}_4}}}{f_1}$$\end{document} 526 Hz396 Hz Table 2Characteristic frequencies of gears at all stages.Characteristic frequencyFormular = 81r = 121Central axis rotation frequency f1 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_1}=\frac{{{v_1}}}{{60}}$$\end{document} 33.33 Hz33.33 HzPlanetary gear rotation frequency f2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_2}=\frac{{{z_1}{z_4}}}{{\left( {{z_3} - {z_4}} \right)\left( {{z_1}+{z_2}{z_4}} \right)}}{f_1}$$\end{document} −16.46 Hz−11.02 HzPlanetary gear rotation period T2 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T_2}=\frac{1}{{{f_2}}}$$\end{document} 0.06s0.09sCycloid gear rotation frequency f3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_3}=\frac{{{{\mathrm{z}}_1}}}{{{{\mathrm{z}}_1}+{{\mathrm{z}}_2}{{\mathrm{z}}_4}}}{f_1}$$\end{document} 0.41 Hz0.28 HzCycloidal gear rotation period T3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T_3}=\frac{1}{{{f_3}}}$$\end{document} 2.44s3.57sSecond stage meshing frequency f2n \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_{2{\mathrm{n}}}}=\frac{{{{\mathrm{z}}_1}{{\mathrm{z}}_3}{{\mathrm{z}}_4}}}{{{{\mathrm{z}}_1}+{{\mathrm{z}}_2}{{\mathrm{z}}_4}}}{f_1}$$\end{document} 641 Hz429 HzFirst stage meshing frequency f1n \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${f_{1{\mathrm{n}}}}=\frac{{{{\mathrm{z}}_1}{{\mathrm{z}}_2}{{\mathrm{z}}_4}}}{{{{\mathrm{z}}_1}+{{\mathrm{z}}_2}{{\mathrm{z}}_4}}}{f_1}$$\end{document} 526 Hz396 Hz Characteristic frequencies of gears at all stages. Changes in the vibration characteristics of the reducer due to various errors Upon analyzing 4,832 manufacturer samples, researchers identified four primary types of faults: eccentricity error, planetary gear defects, forward-reverse noise imbalance, and unsecured fixtures. These defects cause periodic impacts in reducers, particularly noticeable at high speeds. Eccentricity creates uneven meshing gaps, resulting in low-frequency vibrations, whereas planetary gear defects introduce modulation near characteristic frequencies. The forward-reverse noise imbalance stems from asymmetric gear meshing during assembly, leading to abrupt stiffness changes upon reversing. If fixtures are not securely locked, the structural loosening effect is exacerbated, increasing broadband vibration energy, especially in the middle and high frequency ranges. The baseline data for qualified products was established using 4,832 samples, followed by analysis of defective products to determine their characteristic frequencies. By comparing measured vibration spectra with theoretical characteristic frequencies and analyzing natural frequency distributions, the study identifies the dominant frequency components for each fault type, establishing quantifiable vibration-based factory inspection standards for enterprises. Natural frequency calculation Vibration signals often contain the natural frequencies of reducers. To distinguish between natural frequencies and characteristic frequencies, modal analysis was performed on the RV40E 3D model using SolidWorks Simulation software. All components were fabricated from alloy steel with an elastic modulus of 2.1 × 1011N/m2, density of 7700 kg/m3, poisson’s ratio of 0.28, medium mesh quality, and freely divided mesh with approximately 13 mm mesh size. Based on the actual installation configuration of RV reducers, through-holes on the pinion housing were defined as fixed boundary conditions while retaining one rotational degree of freedom for the central shaft. The X/Y/Z coordinate system adopted in finite element modeling perfectly aligned with the experimental accelerometer coordinate system. Meshing details and constraint conditions are illustrated in Fig. 3. The first five natural frequencies and vibration modes of RV40E are calculated and presented in Table 2. Fig. 3Finite element analysis meshing and constraint conditions. Finite element analysis meshing and constraint conditions. Table 3RV reducer initial five modes.Modal orderNatural frequency HzVibration mode front viewVibration mode side viewVibration directionFirst order311 Radial direction of the center axisSecond order507 Axial direction of the center axisThird order2175 Vertical oscillation of pin wheel housingFourth order2258 Horizontal oscillation of pin wheel housingFifth order3805 Planetary gear winding Table 3RV reducer initial five modes.Modal orderNatural frequency HzVibration mode front viewVibration mode side viewVibration directionFirst order311 Radial direction of the center axisSecond order507 Axial direction of the center axisThird order2175 Vertical oscillation of pin wheel housingFourth order2258 Horizontal oscillation of pin wheel housingFifth order3805 Planetary gear winding RV reducer initial five modes. As can be seen from Table 3, the first and second-order natural frequencies mainly correspond to the axial and radial vibrations of the center axis. The third and fourth-order modes correspond to the vertical and horizontal oscillations of pin wheel housing. The fifth-order mode is associated with the winding of the planetary gears. These five modes exclusively occur during first-stage gear engagement, independent of second-stage gear interaction, thus preventing resonance in the second gear. Comparative analysis of characteristic frequencies (f1n) and their harmonic frequencies reveals critical differences: The RV40E-81 model exhibits a 4th harmonic frequency (4f1n=2107 Hz) at 68 Hz below the third natural frequency (2175 Hz), falling within the 1 ± 3% resonance range and posing significant resonance risks. The RV40E-121 model shows a 5th harmonic frequency (5f1n=1983 Hz) 192 Hz below the third frequency, remaining within the 1 ± 9% range. The 6th harmonic frequency (6f1n=2380 Hz) differs by + 122 Hz from the fourth natural frequency (2258 Hz), staying within the 1 ± 6% resonance zone. All frequencies demonstrate high susceptibility to resonance phenomena. Product qualification standards The tests were conducted using the experimental testing platform described in Section Test bench. With a sampling frequency of 5000 Hz and FFT frequency resolution of 1 Hz, all rotational frequencies and meshing frequencies were capped at 526 Hz due to the product’s rated input speed of 2000 rpm. During testing of 4,832 samples, buyers reported resonance bands between 2000 and 2500 Hz, with no post-sale issues occurring above 2500 Hz. Therefore, this study focuses solely on vibration characteristics within the 0–2500 Hz frequency range. After collecting signals from 4832 samples, qualified products were identified based on established factory inspection criteria. Statistical analysis of time-frequency domain waveforms was performed on qualified products, with typical case examples shown in Fig. 4. This dataset selects the test with the largest amplitude during both forward and reverse transitions, extracts 5 s of time-domain sampling data, and applies an FFT transformation to this 5-second signal. Qualified products exhibited highly consistent time-frequency waveforms, serving as a baseline for subsequent evaluations. Key qualification criteria included: stable vibration amplitudes without periodic spikes in time-domain waveforms, with overall amplitudes (xyz) below ±2m/s2; absence of distinct spectral peaks with uniform energy distribution, and vibration acceleration amplitudes (xyz) not exceeding 0.4 m/s2. The qualified standard amplitude was calculated as the vector sum of individual dominant axial components rather than triaxial data, excluding transient amplitudes during commutation and startup phases. Brief amplitude fluctuations during commutation and startup were permitted but required duration under 0.1 s. Experimental data confirmed this threshold applicability to both RV40E-81 and RV40E-121 models. In fault diagnosis, an amplitude twice the baseline value is classified as a minor fault, while three times or more is considered a severe fault. Fig. 4Time-frequency characteristics of the qualified product RV40E-121. Time-frequency characteristics of the qualified product RV40E-121. As depicted in Fig. 4, the qualified product displays the highest amplitude on the Y-axis. The spectrum clearly exhibits the first-stage meshing frequency f1n and its harmonics (from 1 to 6 times), with the 5th and 6th harmonic amplitudes being the most pronounced. According to Table 3, these frequencies coincide with the third and fourth natural frequencies, suggesting resonance. The Z-axis reveals f2n and f1. In the overall spectrum, each characteristic frequency component of f1n is accompanied by a minor modulation sideband, with the sideband frequency centered around the rotation frequency f1. Figure 4 illustrates that the X and Y axes primarily reflect the meshing characteristics of the first-stage gears, whereas the Z axis predominantly shows the meshing characteristics of the second-stage gears. Through extensive comparative analysis of large samples, we have established a benchmark value of 0.4 m/s2 as the frequency standard. According to vibration analysis standards, an amplitude reaching twice this threshold is categorized as a minor fault, while values three times or greater signify a critical fault. Eccentricity fault Excessive eccentricity is a common failure in right ventricular decelerators during manufacturing and assembly processes, which can be effectively identified through vibration detection methods. When eccentricity occurs, it produces distinct periodic noise that is audible to the human ear. The vibration signals are concentrated in the low-frequency range, exhibiting clear periodic impact characteristics. The time-domain waveform displays equally spaced peaks. This type of low-frequency eccentricity error primarily originates from two sources: the eccentricity error of the cam on the eccentric axis, and the eccentricity error of the hole in the cycloidal gear’s eccentric axis. The eccentric axis and cycloidal gear are connected through cylindrical roller bearings. As the eccentric axis rotates, it drives the cycloidal gear to rotate. Each full rotation of the eccentric axis causes the cycloidal gear to roll one pin. Taking the lower cycloidal gear 2 as the reference point, when the cycloidal gear aligns with the two bearing holes, it reaches the eccentric limit position of the entire dimensional chain. At this point, the cycloidal gear positioned on the left is termed the left limit, while the one on the right is called the right limit, as depicted in Fig. 5. The eccentric errors at these two limit positions differ, resulting in two impacts occurring at 180º intervals in the cycloidal gear’s rotational frequency. Fig. 5 The positioning of the cycloidal gear at its left and right engagement limits: (a) left limit and (b) right limit. The positioning of the cycloidal gear at its left and right engagement limits: (a) left limit and (b) right limit. The assembly of RV (Roller Vector) reducers is typically carried out at extreme positions, where eccentricity errors in the eccentric camshaft often manifest along the X-axis. The most severe eccentricity fault arises when the eccentricity errors of the cycloidal gear’s eccentric axis hole and the eccentric camshaft coincide, leading to a non-zero X-axis error and a zero Y-axis error. At this juncture, a distinct stalling sensation can be detected during manual rotation in the assembly process. The eccentricity error of the eccentric axis and the eccentricity error of the cycloidal gear are collinear When the eccentricity error of the eccentric axis and the eccentricity error of the cycloidal gear are collinear, the positional deviation of the cycloidal gear in the X-direction is defined as cij, where i represents the cycloidal gear i and j denotes the hole j, as illustrated in Fig. 6. During machining, each cycloidal gear is marked with a positioning point to ensure accurate alignment, as indicated by the red positioning mark on the cycloidal gear in Fig. 6. During installation, two cycloidal gears are stacked with their positioning surfaces in contact at a 180º angle. The upper circular section on the eccentric axis is referred to as cam 1, and the lower circular section is referred to as cam 2. The radial positional deviation between these two circular sections is eij, where i represents the eccentric axis i and j denotes the eccentric circular section j. Fig. 6Dimensional fit of the cycloidal gear with the eccentric axis. Dimensional fit of the cycloidal gear with the eccentric axis. The two extreme positions of the dimensional chain formed by the eccentric axis and cycloidal gear are shown as the left and right limits in Fig. 5. At the left limit position, the dimensional chain error between eccentric axes 1 and 2 with cycloidal gear 1 is A11, while the error between eccentric axes 1 and 2 with cycloidal gear 2 is A12, resulting in a total error of A13. The dimensional chain relationship at this point is1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ {\begin{array}{l} {{A_{11}}=\left( {{e_{11}} - {c_{11}}} \right) - \left( {{e_{21}} - {c_{12}}} \right)} \\ {{A_{12}}=\left( {{c_{21}} - {e_{12}}} \right) - \left( {{c_{22}} - {e_{22}}} \right)} \\ {{A_{13}}={A_{11}} - {A_{12}}} \end{array}} \right.$$\end{document} When positioned at the right limit, the dimensional chain error between eccentric axes 1 and 2 with cycloidal gear 1 is A21, whereas the error between eccentric axes 1 and 2 with cycloidal gear 2 is A22, resulting in a total error of A23. At this stage, the error of the cycloidal gear remains constant, but the original error direction of the eccentric axis reverses due to a 180-degree rotation. The dimensional chain relationship at this juncture is2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ {\begin{array}{l} {{A_{21}}=\left( { - {e_{11}} - {c_{11}}} \right) - \left( { - {e_{21}} - {c_{12}}} \right)} \\ {{A_{22}}=\left( {{c_{21}}+{e_{12}}} \right) - \left( {{c_{22}}+{e_{22}}} \right)} \\ {{A_{23}}={A_{21}} - {A_{22}}} \end{array}} \right.$$\end{document} The rotation of the eccentric axis results in two periodic peaks caused by the varying limit errors on the left and right sides. The peak exhibiting the higher amplitude is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{max}}\left[ {\left| {{{\mathrm{A}}_{13}}} \right|,\left| {{{\mathrm{A}}_{23}}} \right|} \right]$$\end{document}. The dimensional chain error must be restricted to a predetermined range.3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ {\begin{array}{l} {{A_{11}} \in \left[ {{l_1},{u_1}} \right]} \\ {{A_{12}} \in \left[ {{l_2},{u_2}} \right]} \\ {{A_{21}} \in \left[ {{l_1},{u_1}} \right]} \\ {{A_{22}} \in \left[ {{l_2},{u_2}} \right]} \\ {{A_{min}} \leqslant {A_{13}},{A_{23}} \leqslant {A_{max}}} \end{array}} \right.$$\end{document} where l1 and l2 represent the lower limits of the dimensional chain, u1 and u2 are the upper limits, Amax denotes the maximum elastic elongation limit, and Amin denotes the maximum elastic compression limit. When the total error of the size chain exceeds the elastic deformation limit value, the transmission chain will become jammed, requiring a large torque to pass the jamming point. If the error is excessively large, jamming will occur. When the eccentricity errors of the eccentric axis and cycloidal gear are collinear, the first and second gear meshes simultaneously reach their maximum or minimum eccentricity superposition states. This results in simultaneous XYZ vibration synchronization with identical phase in the time-frequency domain diagram, as depicted in Fig. 7. Fig. 7Collinear eccentricity errors in the eccentric axis and cycloidal gear of the RV40E-81. Collinear eccentricity errors in the eccentric axis and cycloidal gear of the RV40E-81. Figure 7 demonstrates significant amplitude deviations in both x and y time domains. The x-axis amplitude reaches ± 9.5 m/s2, which is 4.75 times the normal range of ± 2 m/s2. The periodic impact is evident with phase synchronization. The frequency domain reveals dominant components at three times the first stage mashing frequency 3f1n, along with modulation sidebands from the central axis rotation frequency f1, and the planetary gear rotation frequency f2. Notably, the X-axis amplitude at 3f1n is three times the normal value, indicating an abnormality. Eccentricity errors in the eccentric axis and cycloidal gear result in radial misalignment of the planetary gear’s axis, causing gear misalignment. This misalignment leads to periodic variations in meshing and axial forces. Parallel misalignment results in radial meshing force fluctuations twice per rotation cycle, (as radial offset induces two axial load variations per revolution. Angular misalignment causes axial force to oscillate twice per rotation cycle, (due to periodic axial component forces generated by tooth surface contact caused by shaft angle. The common frequency of these two phenomena is twice the gear speed frequency 2f1. This characteristic exhibits the highest amplitude in the Z-axis frequency spectrum, indicating severe misalignment in the second stage gear meshing. Unlike conventional ‘constant eccentricity,’ which features a single sideband on both sides of f1n, the gear’s radial runout displays periodic variations, leading to localized, periodic strong impacts. The heightened excitation of higher-order harmonics results in ‘double-frequency modulation’ within the meshing clearance, thus exciting frequencies at 2f1n and 3f1n. Amplitude modulation (AM) patterns emerge around peak frequencies, where the primary meshing frequency f1n serves as the carrier wave while the central shaft rotation frequency f1 acts as the modulation frequency. The spectrum displays side frequency bands (f1n + f1) exclusively on one carrier side, demonstrating single-sideband modulation. This corresponds to the prominent single-side f1 side frequency observed near 3f1n along the X-axis in Fig. 7. The eccentricity error of the eccentric axis and the eccentricity error of the cycloidal gear are not aligned When the eccentricity errors of the eccentric axis and the cycloidal gear are not collinear, the cam eccentricity error of the eccentric axis remains along the X-axis direction, and the cycloidal gear’s eccentricity error along the Y-axis direction is non-zero. In this configuration, vectors cij and eij form an angle θij, with their resultant vector denoted as bij (as shown in Fig. 8). Consequently, the resultant vector force Bi (i = 1, 2) between the upper and lower cams of the eccentric axis forms an angle θi with the X-axis. As θi varies, the extreme positions of the left and right planetary gears may engage with the central axis, causing impact and resulting in a beating phenomenon. The final resultant vector force B, with an angle θ, is ultimately obtained. The cycloidal gear reaches its maximum amplitude when rotating to coincide with B. This amplitude corresponds to the Z-axis signal, where θ represents the phase angle. When the cycloidal gear aligns with the X-axis, the meshing force of the first-stage gear reaches its peak, at which point the Z-axis signal and the X-axis signal have a phase difference θ. Fig. 8The vector sum of the eccentricity errors between the eccentric axis and the cycloidal gear when they are not collinear. The vector sum of the eccentricity errors between the eccentric axis and the cycloidal gear when they are not collinear. Fig. 9Frequency response diagram of RV40E-121 with non-collinear eccentricity errors. Frequency response diagram of RV40E-121 with non-collinear eccentricity errors. A typical case demonstrating non-collinearity between eccentric shaft eccentricity error and cycloidal gear eccentricity error is illustrated in Fig. 9. The z-axis wave peak period corresponds to the cycloidal gear rotation period T3, while the modulation period matches the planetary gear rotation period T2, representing characteristic features of constant eccentricity faults. The frequency domain analysis of x and y axes reveals dominant frequencies at twice the primary meshing frequency (2f1n), accompanied by sidebands f1 and f2. This demonstrates dual-frequency modulation combining amplitude modulation (AM) and frequency modulation (FM), where f1n serves as the carrier wave while f1 and planetary gear rotation frequency f2 act as modulation frequencies. Symmetrical sidebands (f1n±f1, f1n±f2) appear symmetrically distributed around the carrier wave in the spectrum, corresponding to the distinct double-sided bands adjacent to 2f1n in the X-axis direction and synchronized dual-modulated sidebands near f2n in the Z-axis direction observed in Fig. 9. The phase difference between z-axis peaks and x/y-axis peaks indicates non-collinearity with a phase offset θ between eccentric shaft and cycloidal gear errors. Significant beat oscillation observed on the x-axis demonstrates differing vector resultant angles θ1 and θ2 between eccentric shafts, causing phase discrepancies. Comprehensive amplitude analysis reveals severe time-freq9uency domain deviations across the x-axis, resulting in a non-compliant evaluation. Defects of planetary gear The first-level planetary gear is situated at the outermost edge of the reducer. Among the nonconforming products that exhibit excessive noise, the planetary gear issues constitute up to 80%, making it the primary cause of the abnormal noise and vibration in the reducer. Defects in the planetary gear typically present as two distinct problems: (1) asymmetrical tooth surfaces that lead to different noise levels during forward and reverse rotation, and (2) periodic abnormal noises caused by imperfections in the tooth surfaces. Noise difference between forward and reverse rotations The noise difference manifests as a significant disparity in noise levels between the forward and reverse rotation of a reducer, typically exceeding 3dB. This phenomenon primarily stems from asymmetrical gear tooth profiles or the eccentric installation of planetary gears, which causes excessive loading on one side of the tooth surface during meshing. This results in the accumulation of vibration energy in specific directions. The left-right asymmetry of tooth surfaces leads to uneven load distribution during engagement, triggering periodic impacts that become particularly pronounced during reverse rotation. When operating in reverse, the meshing phase lags, causing the engagement starting point to deviate from the designed contact area. This reduces the effective load-bearing surface area, amplifies instantaneous impact energy, and consequently increases the noise amplitude significantly. In practical industrial testing, a noise difference exceeding 2dB between forward and reverse rotation is generally considered abnormal. Such defects manifest in the spectrum as modulation at the planetary gear meshing frequency f1n and its harmonic 2f1n, accompanied by distinct sidebands f1 and f2. The modulation depth intensifies with increased load. This characteristic corroborates the mechanism of eccentric fault formation, further demonstrating that structural asymmetry serves as the primary factor inducing vibration noise. The malfunction becomes more pronounced under low-speed, heavy-load conditions and progressively worsens with extended operation time. During the startup and shutdown phases, the vibration acceleration during gear reversal switching can exceed three times the rated value. Such faults are often undetectable through static inspections in the early stages, requiring frequency domain analysis under dynamic operating conditions. In precision transmission systems, even a slight asymmetry (0.002 ~ 0.005 mm) in the gear tooth surface can cause the meshing frequency f1n amplitude to increase by 10%-20%, or transform higher-order harmonics (2f1n, 3f1n) from ‘undetectable’ to ‘clearly visible’. Figure 10 illustrates the noise measurement process for the reducer. A decibel meter was used to collect noise data from the reducer operating at 2000 rpm, with all noise comparison tests conducted under standardized conditions. Measurement points were arranged according to the General Specifications for Industrial Robot Reducer Noise Testing, positioned on a hemispherical surface 1 m from the housing surface. Average values were calculated after 10-second forward rotation measurements, with reverse rotation tests repeated. Background noise was eliminated using an unloaded background noise subtraction method. Taking a typical product sample with a 5dB noise difference as an example, its time-frequency characteristics during forward and reverse rotations were compared, as shown in Fig. 11. Fig. 10Noise testing. Noise testing. Fig. 11Time-frequency diagram of a 5 dB sample for RV40E-81. Time-frequency diagram of a 5 dB sample for RV40E-81. As depicted in Fig. 11, the time-domain spectrum indicates a substantial disparity in amplitude between forward and reverse rotation. Forward rotation exhibits typical compliant spectra, while reverse rotation displays abnormal spectra with significantly elevated amplitudes. The frequency-domain graph shows normal spectra during forward rotation, but during reverse rotation, the amplitude of the 3f1n harmonic on the x-axis increases from 0.1 m/s2 to 1.3 m/s2, with the amplitude severely exceeding the standard limit. Upon disassembling the gear, an asymmetry of the tooth surface was observed to be 0.01 mm, as illustrated in Fig. 12, and the tooth crest profile exhibited a pronounced inclination. This skewness in the tooth surface results in transient contact line displacement during meshing, which decreases the effective meshing length by about 18%. 3D profilometry measurements verified a stress concentration factor of 1.7 in this area, aligning with the abrupt rise in 3f1n harmonic energy noted in the spectrum. Following the replacement of the gear and subsequent reassembly, the noise discrepancy between forward and reverse rotation diminished to 1.8dB, with higher-order harmonic amplitudes reverting to their baseline levels. Fig. 12Asymmetry in the left and right tooth surfaces of the gear. Asymmetry in the left and right tooth surfaces of the gear. Asymmetric tooth surfaces generally result from machining eccentricity, tool wear, or installation errors. During gear grinding, misalignment between the workpiece’s datum and rotational center leads to uneven cutting depth across the tooth profile. An uneven cutting amount will cause a difference in tooth surface allowance between the left and right sides, thereby forming an asymmetric tooth shape. This may also leave residual hardened layers on the tooth surface, as depicted in Fig. 13. Operational contact stress concentration accelerates wear, further distorting the tooth surface morphology. Unilateral tool wear directly induces tooth profile inclination, resulting in systematic asymmetry errors. RV reducers are commonly used in industrial robots that operate under high-frequency start-stop and variable torque conditions. The dynamic excitation caused by this defect is particularly pronounced during speed changes. Fig. 13Microscopic morphology of the hardened layer on the gear tooth surface. Microscopic morphology of the hardened layer on the gear tooth surface. The asymmetry of the tooth surface causes the meshing position to deviate from the designed contact area. The transition from the non-working zone to the working zone at the tooth tip creates an abnormal excitation source. Excessively small chamfer angles at the tooth tip result in stress concentration. Edge contact tends to occur during meshing, which exacerbates vibration and noise. Edge contact is prone to occur upon engagement, further increasing vibration and noise. As shown in Fig. 14, after grinding, the edge contact phenomenon at the tooth tip is eliminated, leading to a reduction in vibration amplitude, a decrease in noise levels, and a significant attenuation of high-frequency components in the spectrum. Fig. 14Before and after chamfer grinding. Before and after chamfer grinding. Surface defects of planetary gears During the machining of planetary gears, surface defects such as burn marks, impact scars, sintering, and burrs may appear on individual tooth surfaces, as depicted in Fig. 15. When batch warehousing is conducted, these defects can be easily overlooked due to inadequate sampling coverage. The operational impacts at the defect sites generate periodic pulse components in the vibration signals. The frequency characteristics manifest as modulation sidebands related to the planetary gear teeth count, with a center frequency at the meshing frequency f1n and sideband spacing equal to the planetary gear shaft rotation frequency f2. Noise detection reveals distinct periodic hammering sounds. Fig. 15 Gear surface defect: (a) Burns, (b) Impact scars, (c) Impact scars, (d) Sintering, (e) Burrs, and (f) Excessive chamfer. Gear surface defect: (a) Burns, (b) Impact scars, (c) Impact scars, (d) Sintering, (e) Burrs, and (f) Excessive chamfer. Fig. 16Time-frequency spectrum of gear surface defects RV40E-81. Time-frequency spectrum of gear surface defects RV40E-81. Figure 16 depicts a typical time-frequency analysis of gear surface defects. As illustrated in Fig. 9, the sample also displays eccentric faults, featuring a distinct cyclical pattern T3 in the time-domain waveform. The XYZ phases are aligned, with the eccentricity errors of the eccentric axis and the cycloidal gear lying on the same line. The X-axis amplitude exhibits severe deviations, while the 2f1n component is particularly pronounced, and the Z-axis f2n component is clearly discernible. The key distinction is that the side frequencies f1 and f2 in Fig. 16 are markedly prominent, with the amplitude of f1 being particularly noticeable on the Z-axis. This feature is a typical indication of surface defects on planetary gear surfaces. The sideband amplitude is high, and the order is clear and dense, which indicates the significant modulation depth caused by the defect and the robust continuity of fault excitation. Fixture not locked A loose fixture is a prevalent assembly defect in transmission systems. Operational mistakes by workers during detection and installation can cause inadequate preload force on transmission components, resulting in loose connections. This issue can be classified into two typical forms based on the location and mechanism of failure. The primary failure mechanism arises from structural constraint failure-induced rigid body vibration, which occurs when the reducer and fixed tooling (e.g., support brackets, positioning bolt assemblies) fail to reach the designed locking torque. This results in assembly clearance between the fixture and the reducer housing, causing the reducer to become the primary source of vibration through rigid body oscillation. The vibration excitation load originates from two effects: one is the steady-state vibration transmitted by the foundation structure during equipment operation, and the other is the periodic impact forces from tooling collisions within the clearance. This malfunction leads to noticeable radial runout and axial movement in the transmission system, with the X-axis vibration amplitude showing the most significant increase. The vibration frequency peaks at the natural frequency of the fixture-reducer coupling system. The second issue pertains to inadequate coupling between the motor and the reducer, with the core failure mechanism arising from vibrations induced by shaft system misalignment. When the motor output and reducer input, such as couplings or flanges, are not firmly secured, intermittent slips occur during torque transmission, leading to shaft system misalignment. The vibrations are excited by periodic rotational forces caused by this misalignment, including radial bending moments and impact loads from torque fluctuations. The primary vibration frequency in this fault is the shaft system’s rotational fundamental frequency (1x), with harmonic components such as the 2x frequency also present. Fig. 17Frequency plot of RV40E-121 when fixture not locked. Frequency plot of RV40E-121 when fixture not locked. As illustrated in Fig. 17, the amplitude on the X-axis in the time domain significantly surpasses the critical value, and there is also a gradual increase in amplitude. It is evident that the loosening progressively worsens with continuous operation, creating a vicious cycle. The amplitude of 3f1n in the frequency domain on the X-axis exceeds the set limit. Axial vibration amplitudes remain relatively constant, whereas the Z-axis shows concentrated energy within its characteristic frequency band, with the central axis rotation fundamental frequency f1 and its third harmonic component being particularly pronounced. These vibration characteristics closely align with those observed in the case of motor connection failure due to improper coupling. Quantitative fault Indicators Based on a large sample of data, the fault features mentioned above have been classified and summarized. The time-frequency characteristics of the RV reducer display the following patterns: The f2n component consists solely of a first-order component along the Z-axis, which intensifies during eccentric faults.The magnitude of nearly all fault types has significantly increased in the X-direction, necessitating close monitoring of changes in this direction.The Y-axis displays pronounced high-frequency components within the 2000–2300 Hz range, aligning with the reducer’s natural frequency band. The Y-axis primarily indicates loose faults, where the fixture is not securely fastened. The hallmark of loose faults is the generation of high-frequency components, with resonance occurring when the Y-axis frequency components approach the natural frequencies. Specifically, the RV40E-121 model exhibits resonance at 5f1n and 6f1n, while the RV40E-81 model shows resonance at 4f1n.After securing the loose fixture, the friction was eliminated and the amplitude reverted to normal levels. Defects in certain planetary gears, including discrepancies in noise between forward and reverse rotations, burrs, and insufficient chamfering, were rectified through precision chamfering of the gear teeth and surfaces. This resulted in reduced noise levels and the elimination of abnormal sounds. The f2n component consists solely of a first-order component along the Z-axis, which intensifies during eccentric faults. The f2n component consists solely of a first-order component along the Z-axis, which intensifies during eccentric faults. The magnitude of nearly all fault types has significantly increased in the X-direction, necessitating close monitoring of changes in this direction. The magnitude of nearly all fault types has significantly increased in the X-direction, necessitating close monitoring of changes in this direction. The Y-axis displays pronounced high-frequency components within the 2000–2300 Hz range, aligning with the reducer’s natural frequency band. The Y-axis primarily indicates loose faults, where the fixture is not securely fastened. The hallmark of loose faults is the generation of high-frequency components, with resonance occurring when the Y-axis frequency components approach the natural frequencies. Specifically, the RV40E-121 model exhibits resonance at 5f1n and 6f1n, while the RV40E-81 model shows resonance at 4f1n. The Y-axis displays pronounced high-frequency components within the 2000–2300 Hz range, aligning with the reducer’s natural frequency band. The Y-axis primarily indicates loose faults, where the fixture is not securely fastened. The hallmark of loose faults is the generation of high-frequency components, with resonance occurring when the Y-axis frequency components approach the natural frequencies. Specifically, the RV40E-121 model exhibits resonance at 5f1n and 6f1n, while the RV40E-81 model shows resonance at 4f1n. After securing the loose fixture, the friction was eliminated and the amplitude reverted to normal levels. Defects in certain planetary gears, including discrepancies in noise between forward and reverse rotations, burrs, and insufficient chamfering, were rectified through precision chamfering of the gear teeth and surfaces. This resulted in reduced noise levels and the elimination of abnormal sounds. After securing the loose fixture, the friction was eliminated and the amplitude reverted to normal levels. Defects in certain planetary gears, including discrepancies in noise between forward and reverse rotations, burrs, and insufficient chamfering, were rectified through precision chamfering of the gear teeth and surfaces. This resulted in reduced noise levels and the elimination of abnormal sounds. Based on 4,832 valid samples collected in 2025, which encompass all models including RV40E-81 and RV40E-121, the amplitude of 0.4 m/s2 was established as the frequency domain standard amplitude following statistical comparison criteria derived from large-sample analysis. The fault severity was determined by calculating the amplitude ratio between fault samples and standard samples. Statistical analysis of amplitude for various fault types in the aforementioned image samples was conducted, as depicted in Table 4. By calculating the ratio of the maximum amplitude in both time and frequency domains for each fault sample to that of normal samples, quantitative criteria for different fault categories were derived. Table 4Quantitative indicators for fault diagnosis derived from large factory samples.Figure numberFault typeTime-domain amplitude m/s2Time-domain RatioFrequency-domain amplitude m/s2Frequency-domain RatioJudgeFigure 4Normal ± 2 1 0.4 1 Normal Figure 7Sync Eccentric± 9.54.750.92.25Severe faultFigure 9Asynchronous Eccentric± 73.51.64Severe faultFigure 11Noise difference± 2.51.251.33.25Minor faultFigure 16Planetary gear Defect± 73.51.64Severe faultFigure 17Fixture not locked± 840.71.75Severe fault Table 4Quantitative indicators for fault diagnosis derived from large factory samples.Figure numberFault typeTime-domain amplitude m/s2Time-domain RatioFrequency-domain amplitude m/s2Frequency-domain RatioJudgeFigure 4Normal ± 2 1 0.4 1 Normal Figure 7Sync Eccentric± 9.54.750.92.25Severe faultFigure 9Asynchronous Eccentric± 73.51.64Severe faultFigure 11Noise difference± 2.51.251.33.25Minor faultFigure 16Planetary gear Defect± 73.51.64Severe faultFigure 17Fixture not locked± 840.71.75Severe fault Quantitative indicators for fault diagnosis derived from large factory samples. A general principle in the field of vibration diagnosis is that an amplitude ratio of 2 indicates a minor fault, while a ratio of 3 or higher signifies a severe fault. The enterprise has established its own enterprise standards for fault classification by combining general criteria and sample statistics: a time-domain or frequency-domain amplitude of twice the baseline amplitude is defined as a minor fault, and three times or more is defined as a severe fault. As shown in Table 4, when using normal product samples as benchmarks, all fault ratios exceeding 3 can be classified as severe faults, except for Fig. 11 where a ratio of 1.25 cannot be directly determined. Comparing frequency-domain amplitude ratios reveals that all fault ratios surpass 2 except for Fig. 17, with Fig. 11’s ratio of 3.25 qualifying as severe faults. Specific fault types require further confirmation through analysis of fault sample frequency-domain characteristics. In practical factory operations, combining time-domain and frequency-domain amplitude ratios, any single ratio exceeding 3 qualifies as a severe fault. If both ratios are below 3 but one exceeds 2, it is classified as a minor fault. Beyond vibration standards, actual manufacturing specifications must integrate evaluation of other production line parameters. While enterprises adhere to national product standards, their technical requirements typically exceed national benchmarks. A comparison of frequency-domain amplitude values shows that, except for Fig. 17, all fault ratios exceed 2; the ratio in Fig. 11 is 3.25, indicating a severe fault. Conclusion This study concentrates on evaluating the vibration characteristics of RV40E gear reducers across various speed ratios. By conducting a comparative analysis of 4,832 samples, we have established vibration characteristic factory inspection standards suitable for manufacturing enterprises. We have provided detailed, on-site implementable inspection criteria for eccentricity error, planetary gear defects, and manual detection error. These non-conforming product characteristics encompass 90% of the defect types found in factories. The standards are specific, come with sample manuals, and are highly operational. Small batch test verification has revealed that the fault identification accuracy of batch test samples under complex working conditions is 96.7% when applying the evaluation standard from this study, which is significantly better than traditional methods. This research effectively addresses the industry’s lack of clear vibration inspection standards, providing the first comprehensive internal data samples for enterprises. With high engineering application value, it offers data support for the development of the RV reducer industry. The samples selected for this study exhibit easily distinguishable characteristics. In practical factory inspection applications, some cases exist at the edge of the qualification line with ambiguous features. The authors will further supplement the sample characteristics of different fault severity levels under the same fault type in subsequent research, providing fault types and more detailed fault severity information. Frequency analysis requires a higher level of professional expertise from inspectors. To mitigate operational complexity, the research team is developing an intelligent diagnostic module that automatically detects characteristic frequencies and notifies users of abnormal levels. This module, trained using machine learning algorithms, effectively differentiates between noise and fault signals, thereby enhancing diagnostic accuracy. The team intends to pilot this module on production lines to standardize inspection processes and make results visible, aiming to improve both the efficiency and consistency of factory inspections. In manufacturing processes, variations in machining precision and processing errors significantly impact overall performance. This study will focus on analyzing the correlation between machining errors and vibration characteristics, quantifying how different precision levels affect spectral parameters. By developing an error sensitivity model, we will identify critical process control points to provide feedback for manufacturing optimization. Furthermore, through continuous validation of the standard’s applicability using real-world operational data, we ensure the scientific rigor and robustness of detection thresholds. The quantitative evaluation criteria developed in this study have been successfully implemented in enterprises, resulting in a 12.6% improvement in the product pass rate at the factory level within three months. The current standards now encompass routine testing requirements under normal operating conditions, with plans to extend to vibration analysis under complex conditions such as heavy loads and variable frequency operations. The research team will continue to collect field feedback data and dynamically optimize evaluation thresholds to ensure the standards remain applicable and advanced in the long term. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1