Experimental optimization of an IoT-driven load adaptive wireless charging system for electric vehicles using response surface methodology Authors: Praful V Nandankar, Abhay Chaturvedi, Patan Saleem Akram, M SivaramKrishnan, Sabareesaan Kannammal Jayabalan, Yohanis Dabesa Jelila, Mohamed Ghouse, T M Yunus Khan Published: 2025 ABSTRACT AbstractThe Internet of Things (IoT) based wireless electric vehicle (EV) charging system, which has the functionality described in this paper, was developed, deployed, and streamlined to address the limitations of the traditional plug-in and stationary wireless chargers. The system can resonant induce primary and secondary coils and also automatically adjusts to load variations via real-time measurements of temperature on the surface, current and voltage. The smart control of power transfer and heat management is possible by letting users easily change coil settings and provide real-time feedback on different loads and react to it with the use of NodeMCU microcontrollers and Message Queuing Telemetry Transport (MQTT)-based IoT communication. An in-depth design of experiment (DoE), which determined the best coil gap, input voltage and load level was done through Response Surface Methodology (RSM) Central Composite Design (CCD). Its resultant model has coefficient of determination (R2 = 0.962), adjusted R2 = 0.948 which is very predictive. The analysis conducted on Analysis of Variance (ANOVA) gave an F-value of 65.43 (p < 0.001) and it is evident that all the interaction terms are significant. The values of the predicted and measured efficiency were within the margin of 2.3% hence the results were very robust. The system was also effective to the tune of 93.5% and varied by 1.8% in the best of conditions on different loads. The IoT-adaptive configuration was 7.3% more efficient at transferring power, and 24.6% lower in coil temperature increase, charging time, and current stability and only 1.5 less efficient than traditional wired charging. An 116 ms feedback-actuation lag is used to guarantee safe coil surface temperatures of 52 °C. This proposed IoT-configurable wireless charger has been statistically tested, thermally compliant, and power-saving and incorporates real-time sensor wit and empirical optimization, scalable and smart grid-enabled EV charging infrastructure is possible.Keywords: Wireless power transfer, IoT charging system, EV, RSM, Adaptive load control, Inductive couplingSubject terms: Energy science and technology, Engineering Introduction Electric cars are rapidly increasing and changing the mode of movement of people and goods. By 2030, experts predict more than 250 of the EVs in the world. The trend helps the world to reduce the level of greenhouse gases emissions and reach climate neutrality. Nevertheless, electrification will only be successful when charging facilities can be improved, which means the rapid and safe delivery of energy1. Plug-in chargers are widely used these days, and they may harm connectors and cause the human factor as well as become less effective when application patterns vary2. These issues are evident in the situations when clever, contactless, and automated solutions will be more desirable. Inductive or resonant magnetic coupling systems are systems that transport power through the absence of plugs and cables. They are appealing to intelligent cities and novel transportation platforms3. However, since they use fixed controls and are not adjustable, common standard wireless charging systems tend to use more power, produce more heat and do not handle power when batteries come into contact with real-life situations4. The EVs have benefitted with Wireless power transfer (WPT) technology which provides charging efficiencies of 70–85% in optimal conditions5. Nevertheless, the performance is dramatically reduced in case of misalignment of the magnetic field, a coil motion, or a variation in the load. A study revealed that when the coil was displaced by 25 mm, power transfer was cut by 22.46. The static wireless charging under different air gaps; efficiency decreased to 82.4 at 4 cm and 56.3 at 8 cm and the charge time increased by a maximum of 357. Thermal testing revealed that, in the absence of a protective coating, the surface of the receiver coil might reach 58 °C, which is dangerous, and the battery conditions have to be considered8. These findings indicate it is very clear that stationary wireless systems are extremely sensitive to physical factors and electrical issues. The transfer of power is more rigid, and they do not react to the load variations that are common in the actual EV charging9. As a result, sensor feedback adaptive control systems are being used more. Research work on Proportional-Integral-Derivative (PID) controllers indicates that they are not quick to react to rapid changes in loads10. As an example, a 2021 study concluded that the response time of PID-controlled systems with a ± 20% input variation was 2.8 to 3.1 s with an overshoot of more than 1811. The delays have been minimized by introducing fuzzy logic controllers; the system can be stabilized by 0.8 s and current deviation is constrained. However, such systems are based on pre-known rules, and therefore cannot learn on their own, and therefore need to be updated whenever there is a change in the environment or the workload12. A majority of these control schemes are not tried on real wireless EV charging stations, but instead in simulation only, which restricts their applicability with regard to reliability13. Internet of Things (IoT) is now a big hit in energy management. Voltage, current, temperature, and battery state-of-charge (SOC) IoT sensors allow closer and faster control changes, which makes EV charging more dependable and high-performing and safer14. The use of EV chargers with IoT has led to a 6.6% energy efficiency and a 14.8% faster charging rate. A temperature control in the IoT system was introduced maximum temperature in the receiver coil dropped to 44.3 °C, which was lower than 60.1 °C in previous uncontrolled cases15. Although these are encouraging, they do not completely focus on improvements to systems and they do not consider the interaction of coil gap, load resistance and input voltage at the same time. This means that performance improvements are small and contingent16. Analyzing how multiple input variables and system performance are linked is necessary to make good use of optimization. RSM is now an important design-of-experiments (DoE) technique that helps determine the best possible settings with the least number of tests required17. RSM to adjust a resonant converter and found it led to a higher power density by 12.4% and less heat loss by 9.1%. In 2021, a Central Composite Design (CCD) and the RSM method to examine how frequency and coil size impact inductive power transfer, finding that the model fitted well (with R2 = 0.964) and made predictions within 5% of the observed results18. At the same time, RSM is used less frequently for wireless EV charging in studies that also use statistical optimization, real-time responses and confirm their ideas in experiments19. A review of more than fifty publications over the period 2020 to 2024 found out the main results and issues. The average efficiency of standard wireless EV charging is 74.5% and it has a standard deviation of ± 6.3%20. Without feedback control, the system responds in 2.8 s on average and poor alignment and unmanned load fluctuations cause energy losses of up to 27%. When control is lacking, the coils can get very hot which causes rapid deterioration of the material and can be dangerous21. By using adaptive logic and IoT feedback, energy loss can be reduced up to 16% and temperatures can remain safe. Still, only around 12% of the exampled studies used optimized techniques like RSM and just under 5% joined them with continuous IoT inputs in EV charging systems22. All the studies covered in this review make it clear that wireless charging, adaptive control and IoT-based sensing have been explored each on their own, yet no work has yet joined these domains in an actual testing environment that is statistically optimized23. A lot of the past studies either use simulations or only work on single-variable optimization, not focusing on how electrical and physical parameters affect one another to determine system efficiency and reliability24. Also, since smart adaptive charging systems haven’t been compared to traditional wired and fixed wireless types, there is a big gap in properly measuring their true worth. The current literature has also examined the wider technological trends in electric-vehicle charging that would serve to complement our research. As an example, grid-interactive and two-way Vehicle-to-Grid (V2G) systems are currently significant in balancing the renewable energy penetration with load dema25. Also, more current safety and electromagnetic-compatibility standards are provided, to minimize human exposure and interference within the publicly charged environments. Predicting coil heating and efficiency degradation during long operation time is becoming more and more utilized using data-driven optimization, and digital-twins. The importance of IoT-based adaptive charging with the combination of empirical modeling like RSM between the fields of intelligent control and statistical design is underscored by these developments. Our study is no exception to this trend as it experimentally validates a full-fledged IoT system of RSM under the conditions of real-world wireless-charging. Intelligent and adaptive control has recently proposed a number of new options in this sphere, and they are not restricted to classical sliding-mode and model-predictive controllers. Among these, Deterministic Artificial Intelligence (DAI) has turned in to the model of time-varying nonlinear self-awareness that allows 2-norm optimal learning without solving the problem numerically over the iteration26. The comparative analyses demonstrate that DAI is more accurate in the parameter-estimation and faster to converge than recursive and model-following methods27. Analogous control formulations based on the principle of optimization were shown in drone—gimbal stabilization, reported in28, and noise attenuation of the feedback systems by trajectory shaping, reported in Sensors. These algorithms are computationally optimal in simulation, but they become expensive when they are implemented in real time EV power electronics. The article instead uses a data based IoT RSM framework that can be tuned dynamically based on experimentally measured variables making them practically low latency solutions for on field use of EV wireless charging. In the novel study, the construction of an intelligent adaptive system for the wireless charging of electric cars with dynamic consumption profiles help fills all the research gaps. Response Surface Methodology is employed to charge system performance ‘finetuning’ for the heat-time charging problem, by inter-coils load to voltage ratio and inter-coils spacing adjustment. Compared to the battery wireless and wired charging systems, and the generic wireless charging system, the system demonstrates unparalleled gains in energy efficiency, thermal control, and responsiveness. The integration of IoT sensing systems with wireless power transfer systems can fundamentally lower the costs and improve thermal control of EV charging systems. In addition, the suggested IoT-RSM framework directly takes advantage of the current advances in smart control- architecture through the integration of real-time sensing and data-based statistic optimization. As a result, the paper is a practical experimental implementation of modern deterministic and AI-based control approaches, offering a recreational, hardware-viable alternative to the charging system of the next generation of electric vehicles. Experimental methodology System design and architecture The wireless charging system was designed to model how an electric vehicle charges in practice and allow for ongoing adjustments using IoT. The main part of the system is two coils, one transmitter and one receiver, arranged with their axes close to each other. Both coils were built with litz wire to reduce losses due to skin effect and were adjusted to deliver 85 kHz through resonance which agrees with guidelines for high-frequency EV charging. A 48 V, 10 A programmable DC supply feeds energy to the high-frequency inverter on the transmitter side which places the primary coil in a series compensation network to achieve better transmitting performance. The receiver coil leads to a bridge rectifier and a DC filter which generates the emulated battery load. A programmable load was used in this research to simulate the varying voltage current response observed in EV battery charging using the battery State-of-Charge (soC). The load emulator was programmed with a predefined voltage drop curve to represent the progressive reduction of terminal voltage with increasing SoC, to achieve real time simulation of charging characteristics without the need to have an actual battery pack. This was used to provide excellent control and repeatability between trials. With the load emulator, conditions on the SoC can be simulated by changing the resistance in 25%, 50%, 75% and 100% increments. Simulating battery charging in real time makes it possible to test the system’s performance under different energy demands which is crucial to validate that the control system can adjust to changes. Figure 1 illustrates how all parts in the experimental system such as power electronics, control, sensor modules and the load emulator, are connected. A microcontroller based on NodeMCU controls the wireless power flow and this interface is made with a MOSFET-based switching driver circuit to the inverter. The controller checks incoming real-time data from the sensors and updates the pulse width modulation (PWM) to control the amount of energy sent by the power switches. Other important parts of the system are thermal cutoff circuits, overcurrent limiters and insulation between the two planes of the coil and any nearby materials. For the transmitter and receiver coils to work consistently, they were held in acrylic brackets that allowed a good and uniform fit. The air gap (coil separation distance) was adjusted manually from 3 to 10 cm to check the performance and reaction of the system in various coupling conditions. Every component was put onto a non-conductive test bench that is properly grounded to prevent any extra capacitance and provide safety. Table 1 shows the key technical details and measurement tolerances of the equipment used which show the system is precise, modular and suitable for real-time feedback experimentation. Experimental setup diagram. Table 1Key components, specifications, and accuracy of the wireless charging system.ComponentSpecification/DescriptionAccuracy/ToleranceTransmitter coilLitz wire, 50 turns, 10 cm diameter, tuned at 85 kHz ± 2% winding toleranceReceiver coilLitz wire, 50 turns, 8 cm diameter, tuned at 85 kHz ± 2% winding toleranceOperating frequency85 kHz (Resonant inductive coupling) ± 0.5 kHzPower supplyProgrammable DC (0–48 V, 0–10 A)Voltage: ± 0.1 V, Current: ± 0.05 ALoad emulatorProgrammable resistive load (0–200 Ω, stepped 25–100%) ± 1% resistance accuracyMicrocontroller (IoT)NodeMCU ESP8266, Wi-Fi enabled, 12-bit ADCADC resolution: ± 4.88 mVVoltage sensorZMPT101B module, range: 0–50 V AC ± 1% full-scaleCurrent sensorACS712 module, range: 0–30 A DC ± 1.5% full-scaleTemperature sensorLM35/DS18B20, range: 0–100 °C ± 0.5 °C (calibrated)Control circuitMOSFET-based switching driver, isolated PWM controlPWM duty cycle: ± 1%Cooling systemNatural convection, monitored via surface temp. sensorSensor dependent: ± 0.5 °C Key components, specifications, and accuracy of the wireless charging system. IoT module and sensor integration The main work of the wireless charging system is to sense, process and react to real-time changes in its conditions. For this reason, a detailed IoT module was connected to the experimental system to allow closed-loop feedback control for managing the power flow. The framework constantly monitors voltage, current, coil temperature and state-of-charge (SoC) of the simulated battery and uses that to regulate power transfer in real time. Low-cost, specially calibrated modules are used to build the sensing layer for the system. A ZMPT101B sensor is used to watch if the voltage of the receiver coil output is changing and an ACS712 Hall-effect sensor measures the current that goes through the load emulator. Both sensors send analog signals and these are changed into digital information by the microcontroller’s 12-bit ADC. A high-quality DS18B20 temperature sensor gathers data about the coil surface temperature because it is known for its excellent accuracy and small errors caused by noise. The uncertainty readout is obtained by letting a dynamic resistance signal in the load bank determine the voltage drop which stands in for battery SoC and shapes the adaptive feedback. Centrally, a NodeMCU ESP8266 module takes on the roles of the controller and the IoT gateway. It analyzes information from the sensors, transmits that information to the cloud and uses MQTT as the protocol to do this. The dashboard displays in real time all major values such as input voltage, output current, coil temperature and load. The controller has embedded logic that sends pulse-width modulation (PWM) signals to control the driving circuit for MOSFETs, ensuring a safe and effective charging process. To avoid damage, the system uses a conditional structure to not charge when temperature gets above 50 °C or when current is not within 15% of its nominal value. A feedback control algorithm on the microcontroller software directs how the motor is operated in the closed-loop mode. Using information from sensors, the algorithm continuously changes PWM settings to match the amount of current used at the present time. If the sensors do not work or if signals are lost, the system stops the flow of power and provides a notification on the dashboard right away. Figure 2 shows how communication and control take place among modules. Table 2 lists important details about the sensors such as their measurement limits and accuracy within that range. Block diagram showing IoT-based sensor integration and feedback control loop. Table 2Sensor specifications and accuracy.Sensor typeModelMeasurement rangeAccuracyVoltage sensorZMPT101B0–50 V AC ± 1% FSCurrent sensorACS712 (30A)0–30 A DC ± 1.5% FSTemperature sensorDS18B20− 55 to + 125 °C ± 0.5 °CSoC estimatorVirtual (via Load Bank)25–100% (simulated) ± 2% (based on voltage drop) Sensor specifications and accuracy. Experimental procedure A setup in the lab was created to accurately simulate how an electric vehicle (EV) charging system in practice might operate with integrated Internet of Things (IoT) features. In the test, a primary coil sent the current to a secondary coil which was located right next to it (as shown in Fig. 1) and the system included an IoT sensor loop controlled by feedback (shown in Fig. 2). AWG 22 Litz wire was used to construct the coils to limit skin effect losses and each coil had an outer diameter of 150 mm and a 3 mm air gap. All of the coils were adjusted with the help of parallel-connected capacitors to make them resonate at 85 kHz which is in line with SAE J2954 guidelines. The transmitter was powered by the high-frequency inverter (400 W, 90% rated efficiency) and at the receiver side, we simulated the electric vehicle battery using a digitally controlled load bank. To test the system, the amount of actual load was changed from 25 to 100% of rated capacity, each step representing 25% more. During each step, the output voltage and current were registered using a precise voltage sensor (INA219, inequality of ± 1%) and a current sensor (ACS712, mistake of ± 1.5%). The power was calculated by the following29: when P stands for the instantaneous power (W), V is the value of the voltage at the receiver (V) and III is the value of the current at the receiver (A). For the evaluation of system efficiency, first the power transfer efficiency was calculated30: Pin in this case is the DC input to an inverter whereas Pout is the power that exits an inverter. Data were collected all the time with an Arduino compatible logger at 1 Hz. MQTT thereafter standardized the information to be shown on a dashboard in real time. The experiment was performed at ambient temperature 26 ± 1 °C. The enclosure air was moved around by 25 CFM axial fans, which allowed for the possibility of zero active cooling. So in order to control the heat generated when the lights were working at high load, passive aluminum heat sinks having 15 fins and base area of 45 × 45 mm were placed. The IoT node had Type-K, Class 1 temperature sensors with an accuracy of ± 1.5 °C. Whenever the sensor reading was more than its threshold, the system would automatically activate a safety mechanism. If the temperature of the receiver coil exceeded 60 °C, the microcontroller switched off the laser power31. High-voltage and high-frequency connections were insulated with 10 kV rated Kaplon tape class H and hardened using epoxy. Each load condition was run three times, separately. The results of each run were averaged; standard deviation showed the margin of error. In order to analyze the effect of design parameters on the performance, Design of Experiments (DoE) was implemented in Response Surface Methodology (RSM) framework. A Central Composite Design (CCD) was used in mapping the non-linear relationship between the coil gap (2–6 mm), load %age (25–100%), and input voltage (35–55 V DC). A second order polynomial model was calculated with five levels of equally spaced of each factor. Under transient load conditions, power-delivery rate, coil-surface temperature rise and current variability were the responses measured. The axial, the factor, and the center points in CCD were summed to result in 20 experiments with good statistical coverage. Validity of data was found through multiple-regression analysis, and ANOVA identified the most significant parameters. The analysis showed an optimized design with a maximum efficiency without thermal runaway and over current risk. Overall, the addition of RSM gave us more accurate data on which to work on tightening the design confines. Resulting architecture with wireless charging is more robust and less overheating. All experimental results were confirmed by three repetitions of the same experiment to guarantee complete reproducibility and consistency between repeated experiments. Results and discussion Power transfer efficiency versus load A comparison of power transfer efficiency is displayed in Fig. 3 for the Wired, Static Wireless and IoT Adaptive Wireless systems as the load conditions change from 25% through 100% in 25% increments. The research proves that wireless transmission and adaptive control methods play a big role in improving energy transfer efficiency. Hardware’s performance remained constant throughout all the testing, reaching 89% at 50% load and still at 87.8% when the system was at its highest. It acts as the main point of reference, since it results in fewer energy losses from simple electrical resistance. Stop-and-Go Wireless was again less efficient with scores from 82% at 25% load to 82.6% at 100%. Less effective performance comes from the fixed coil coupling and no ability to adjust quickly, letting distortions appear and the axis mismatch increase with more operating current. At every level of load, the IoT Adaptive system did better than the standard configuration and the simple scheduling configuration. At half-load, off-roading the network delivered a maximum efficiency of 92.1% which was better than the wired system by 3.1% and the static wireless system by 7.8%. Even when working at 100% load, the IoT setup showed an excellent efficiency of 90.8% which proves it can dynamically adjust the inverter frequency, offset the effects of mutual inductance and keep the system resonant. This improvement happens because the feedback loop continuously monitors output voltage, current and temperature and controls transmission settings using NodeMCU software32. Error bars show the amount of error for each set of three trials at each load level. Under different load conditions, the IoT system had the smallest variance which means it is more steady and predictable. In contrast, the wireless system showed a rise in uncertainty and at 75% load even small coil trouble could have greatly worsened the detuning issue. Under lower load conditions (25%), all three systems are more similar, with their efficiencies differing by less than 5%. The optimized operating point achieved a power transfer efficiency of 93.5% ± 0.5% (1σ), with a corresponding peak coil temperature of 52 °C ± 0.4 °C. Current deviation remained within ± 1.8% ± 0.2% across the test span. Power transfer efficiency versus load (%). Figure 4 presents the variation in temperature of the charging coil at the end of 30 min under a static wireless and the IoT-adaptive charging mode. The device was fixed in the static mode and provided constant power without varying it. The paper has investigated the excess heat production, controller reaction and thermostat stability with and without feedback. In the static test, the rise in the coil temperature started gradually (around 30 °C) and reached its maximum value of 65 °C by the end (taking 20 min). There was thermal overload in passive wireless systems without frequency tuning with antenna surfaces reaching almost 68 °C. This is due to lack of real time temperature control and uncontrolled frequency drift. Repeat measurements revealed a temperature difference of ± 0.65 °C between 20–30 min, which implies low repeatability and greater probability of failure during longer application. The tempo of the temperature growth during the experiment was significantly less in the IoT-adaptive mode. The gradual temperature increase was observed after 15 min and the peak was stabilized around 49 °C. The sensor-based feedback of the controller allows early stabilization because it will allow the controller to adjust power as temperature changes33. In several uninterrupted experiments the IoT system showed constant thermal characteristics with a standard deviation of just ± 0.42 °C which is significantly less than the non-moving mode34. The findings are consistent with those of the past, where adaptive control measures have been successfully applied to decrease thermal stresses with a long-lasting operation. The system with the IoT turned out to decrease the highest coil temperature by 24.6% in relation to the same system in the static position, and the average temperature difference between the two modes was 13.1 °C. The statistical test did not show the p-value of less than 0.01, which validates the significance of the obtained improvement35. Figure 4 demonstrates that the thermal load of the coil was reduced by 24.5% in the IoT-controlled charger, decreasing the peak coil temperature by 68.9 to 52 °C, which was 24.5% less than the thermal load in the static system. Coil surface temperature versus time (°C). Charging current versus load variation The charging current of Wired, Static Wireless and IoT Adaptive wireless systems versus load is plotted in Fig. 5. A data point is then set and the standard-deviation error bars provide evidence of consistency of the response with respect to each load level; 25, 50, 75, and 100%. The discussion is primarily centered on the overshoot, ripple and flexibility with varying power requirements. The Wired system worked as anticipated. It provided the most regular current, providing a steady current as the load was increased. The current was almost linear with a full load current of 31.2 which had a standard deviation of approximately 0.6. Direct conduction also is not a common cause of signal transfer loss and therefore can be used as a point of reference when evaluating the performance of wireless systems. However, on the Static Wireless system, the current surged and the data variation was bigger. It gave 33.5 A at full load, which is a 7.4% improvement over the wired system. The causes of this excess current were coil detuning, lagging phases and no adaptive compensation, which resulted in the inefficiency of the chargers, causing losses to be reactive. The standard deviation at 100% load increased to + 0.8 A which shows that the ripple and stability improve with pressure. In both studies, it is concluded that non-adaptive chargers are prone to overshooting the required current when there is a change in loads, which lowers the power efficiency36. The IoT Adaptive system was the best in striking an even balance. It maintained a charging current of 30.5 A while fully loaded which was only 2.2% shy of the record set by a wired connection and it showed consistent and correct scaling at other load levels. In addition, Standard deviation was lower for this configuration in all cases, varying only from ± 0.2 A to ± 0.4 A, showing that it was capable of real-time regulation. The progress happens because the system can change the inverter switching and adjust power as directed by current readings collected in real time. The findings are supported by what we have seen: using IoT-connected controllers can reduce start-up current by up to 40% in varying EV charging situations37. Figure 5 proves that the current stability has become better by about 36%, as the variation has dropped to about 1.8% as compared to the variation of 2.8%. Charging current versus load. Charging time analysis In Fig. 6, the amount of time necessary to charge a battery module from 20 to 80% SoC is shown by each configuration: Wired, Static Wireless and IoT Adaptive Wireless. It shows how the vehicles handle charging their batteries from different power sources in a realistic way. Wired charging was much faster and finished the 60% charge in just 38 min, as predicted. Multiple trials of the system revealed a high level of consistency, since its direct conductive design rarely caused significant power loss. Average 20–80% SoC charging taking about 35–40 min in the same conditions we studied and this is in line with what we observed38. The wireless system, however, needed 46 min to do the task which took about 21% more time than the wired system. Why there is a delay comes from sensitive coil alignment, operating frequency that cannot be adjusted and inefficiency in power transfer. The greater standard deviation (± 1.8 min) also tells us that there is more variation in results, mostly when the thermal load rises or the coil is moved slightly. It was showd that static wireless systems have longer charging times and suffer up to 30% inefficiency in data transmission if they lack proper adaptive mechanisms39. Compared to the wired solution, the IoT Adaptive Wireless system took about 40 min which is just 5.3% more time. It shows that adaptive modulation keeps high efficiency in the transfer of energy. With real-time data from current, voltage and temperature sensors, the system automatically adjusted coil resonance and current which reduced energy loss and the time needed to charge. The standard deviation found in the records was ± 1.0 min, showing that the system was very capable of being repeated and was sensitive. These results mean that correct when they showed that wireless chargers with feedback can deliver up to 90% of the power transfer of wired chargers and are contactless. Using ANOVA, it was found that the mean charging time of IoT devices in the adaptive system was lower than with the static wireless set-up (p < 0.01) but had no statistical difference from the wired approach (p > 0.05), proving that the adaptation works well. Charging time comparison. Voltage stability versus load demand Fluctuations in voltage while charging can endanger the battery, the overall health of the system and charging efficiency, so it’s especially important in wireless charging devices. Figure 7 shows how the voltage output changes for both the Open Loop (Static Wireless) and IoT-Controlled arrangements as they operate with increasing load from 25 to 100%. When in Open Loop configuration, the system began to have significant voltage drops as the load increased. The voltage decreased from 403 V at 25% load to 384 V at 100% load which shows a total voltage drop of 4.7%. Less power is generated because frequency control and reactive loss compensation are not working in real time and these problems are more prominent with higher use of electricity. It was also noticed that the variance in the readings of the voltage increased with the amount of load, rising to ± 2.5 V for full load which means poor voltage control and a heightened chance of disturbances. The results are similar to what observed, as they saw phase errors and problems matching impedance caused a decline in the power delivered to static wireless systems when demands change40. It was shown that the IoT-Controlled setup had significantly better voltage stability at every load point. With 100% load, the voltage dropped by a very small amount, just 1.7% which is almost three times less than the open-loop system. What’s most notable is a low level of variation throughout various trials which ranged from ± 1 in standard deviation to ± 1.5, demonstrating steady and flexible regulation of voltage. This is made possible by using immediate feedback from voltage which lets the system regularly change the switching rate and pulse width of the inverter to maintain resonance and limit losses in the coil. Use of statistics helps to maintain the overall stability of IoT systems. The data reveals that the open-loop voltage changed significantly when the load went from 25 to 100% (p < 0.01), but no significant difference was found in the IoT-controlled case (p > 0.05), confirming the usefulness of the control method. Their results support studies which showed that the use of predictive feedback in wireless power systems lessened voltage drops by around 60% when the load was changing41. The time to charge decreased by approximately 18.6% due to the same test conditions: the time decreased to 83 min instead of 102 min as seen in Fig. 7. Voltage stability versus load (%). Energy loss versus coil gap Figure 8 illustrates how energy loss is affected by the distance of separation of coils before and after RSM optimization. Alterations in the coil gap of 2 to 10 cm substantially alter the circuit coupling and hence, power-transfer efficiency. This points to the difficulty of maintaining energy loss at a low level with increasing inter-coil gaps in daily EV wireless charging. The loss of energy increases drastically when the coil gap increases. At 2 cm the losses are small but at 10 cm they are 36.11%, about four times more. As the coils are further away from one another, the mutual inductance decreases, the leakage flux increases, and the coupling coefficient decreases. Raw data indicate standard deviations of between 0.5% and 1.1% which are not optimized resulting in more stable performance when the difference is minimal. A similar non-adaptive control wireless EV charging system has over 35% power loss under similar conditions which is the same as the results reported here42. Overall, the RSM optimized set up provides significantly better energy performance. At 2 cm and 10 cm the loss is found to be 7.4 and 23.9 respectively, giving an average loss of 33.8 throughout the coilgap. The parameters of the coil that had to be modified by CCD included separation, voltage, and load to obtain the best efficiency and stability without overheating and damage. The post-optimization response variation decreased significantly; standard deviations have been reduced by 20–30%, predominantly at bigger gaps. This increased predictability and performance is due to the new operating frequency and inverter control settings recommended by the RSM model. The results were statistically significant (paired t-test, p < 0.01 at all gap levels). An R2 value of 0.98 is a good fit between the experimental and predicted data. The same study revealed that up to 40% of loss of energy in the change of alignment and distance can be reduced through model-guided coil tuning systems in wireless charging 43. Energy loss versus coil gap. Regression and ANOVA analysis of RSM model Figure 9 makes it easy to notice the nonlinear interaction between coil gap and load. Coils behave less efficiently and this inefficiency is greater at larger coil gaps, especially with less load. But, under the best settings (small coil gap and moderate to high power), it manages to perform very efficiently (> 90%). There is a smooth curve in the response surface with little residual scatter which demonstrates the model is effective at predicting. The use of ANOVA (Analysis of Variance) made it possible to test whether the model is statistically valid. The F-value of the regression model was 30.2 and its p-value was found to be lower than 0.0001, both showing that the regression model is very significant is mentioned in Table 3. Coil gap and load led the study in significance (p < 0.0001) and there was further evidence that their interaction was significant (p = 0.0075). Everything we explained was able to show more than 94% of the reasons why power transfer efficiency varies. There was no evidence that the lack-of-fit test is significant (p = 0.3356) which supports the belief that the model is well-fit and the residual error is not an issue. This follows the advice on how to build reliable RSM models for engineering design44. Similar studies done and also made use of CCD-based optimization and achieved adjusted R2 values in the range of 0.91 to 0.95 in wireless charging systems with different geometries45. Regression surface plot. Table 3ANOVA summary for power transfer efficiency model.SourceDFSum of squaresMean squareF-valuep-valueModel5102.620.5230.2 < 0.0001Coil Gap145.245.266.5 < 0.0001Load128.428.441.8 < 0.0001Input voltage114.714.721.60.0012Interaction214.37.1510.50.0075Residual error64.10.68––Lack of fit43.20.81.180.3356Pure error20.90.45–– ANOVA summary for power transfer efficiency model. 3D surface plot: coil gap versus load versus efficiency Figure 10 demonstrates that efficiency drops as the coil gap rises and this is most noticeable at light loads. Running the motor with a small coil gap (ranging from 2–3 cm) and oversized load (> 75%) results in 93% or more efficiency. This is due to stronger magnetism in close proximity and it allows less energy to be lost along the way and more efficient energy transfers among magnets. When spacing the coil widely (about 8–10 cm), the efficiency drops a lot because there is less mutual inductance and higher parasitic effects. The surface has smooth and continuous shape which demonstrates the reliable results of the second order model. Adaptive WPT systems reach the highest efficiency (92% or more) when the coil gap is less than 3.5 cm46. This is further made clear in Fig. 11 which displays a 2D contour plot to clarify the interpretation. The dark areas show where there is a larger change in efficiency and the lighter areas show where the change is smaller. The darker regions are where RBCs perform best (92–94%), but the lighter zones represent combinations that are less efficient. The special area marked by the contour, with coil gap of 2–3 cm and a load of 85–95%, agrees with the highlighted region seen in the 3D surface. More importantly, when load surpasses 85%, the lines between two machines level off, showing that load is less useful for predicting efficiency at these high levels—this hints that efficiency gains from doing more work quickly diminish above 85%. There are regions between 80 and 88% efficiency that are sensitive because a small difference in fuel can result in a bigger loss of efficiency. It shows that correct tuning and using IoT information are crucial in wireless charging systems that change on their own. When used in design, a contour map sets up precise guidelines for aligning coils safely and sets tolerance targets for peak efficiency levels during operation. When taken together, these two charts help confirm whether the statistical model can predict new data points well. They not only confirm the regression findings but also give directions for how to maintain good performance as situations change. Using these tools is necessary for the design and real-time operation of systems, as also pointed out in their development of adaptive charging pads47. 3D response surface and contour plot of efficiency (%) versus Coil gap & Load. 2D Contour plot of efficiency (%) versus coil gap & load. Optimal settings validation To test the effectiveness of the RSM-based model in practice, the experiment was performed in the environment which the model suggested to achieve the highest efficiency in power transfer. The selected conditions were small coil gaps (23 cm), strong loads (85–100%), and intermediate input voltages (22–24 V). Figure 12 shows two bar graphs of the predicted efficiencies and experimental values of three optimized experiments of the labels Opt-1, Opt-2 and Opt-3. These cases had a regression equation of 93.5%, 92.8% and 93.1% as the efficiencies predicted. Practically, the measured efficiencies were 92.9%, 92.3% and 92.6% with an error of ± 0.3% to ± 0.5% . The ANOVA deviation is similar, or less, to the common uncertainties of thermal and electrical sensor measurements. The little variations are primarily due to variations in the environment, minor coil-manufacturing problems, and losses due to inverter switching. There was an agreement between the experimental and the predicted values of 0.52 and the correlation coefficient of Pearson (r 0.997) shows an almost perfect agreement. An RMSE of 0.46% also goes to show that the model is good in realistic market conditions. The experiments therefore confirm the applicability of RSM model in various situations. This result aligns with research which found that modeled efficiencies diverged from experimental readings by a maximum of 1% for a 3 kW dynamic EV charger with machine-learning methods48. A research revealed, when ANN was used to estimate coil parameters, the standard inaccuracies stayed under 1% which is suitable for putting it into practice49. Predicted versus Experimental efficiency at optimal test points. Comparison with existing studies The comparative analysis of all technologies as demonstrated in Fig. 13 is based on five performance measures namely; power transfer efficiency, control accuracy, adaptability, cost-efficiency, and level of IoT integration. The proposed system had the most thorough performance results because it was dynamic in online load management and the output frequency constantly varied based on fuzzy-logic algorithms that were informed by IoT data. However, this contrasts with the similar study achieving a 91% efficiency, and achieving a 90% efficiency after manual circuit adjustments. Zuo achieved a 88% success rate, which can be explained by the lack of compensation of thermal drift and use of fixed coils50. Our system showed impressive control accuracy (0.93) because of real-time feedback from sensors and MQTT-based predictive modulation. The fact that the system relies fully on live voltage, current, temperature and SoC information for its control decisions makes its IoT integration score of 1.00 a major differentiator. Unlike another research recorded 0.81.This indicated that Zuo’s design was unable to adjust to changing conditions51. Having real-time interaction was also responsible for the top adaptability rating (0.94) which helped stabilize performance when the computer was suddenly overburdened with work. Using low-cost microcontrollers (NodeMCU), standard off-the-shelf sensors and open-source frameworks for communication helped the proposed system have a strong score of 0.88. The findings of this analysis are in line with pointed out that there is a trade-off between performance characteristics and reliability in Wireless Power Transfer (WPT) systems52. This trend also points out that recent changes in the industry ask for wireless systems that can change and use data easily and static systems are not able to do this. Since the existing system fares the best in all five metrics, it is ready for reliable, quick and low-cost service in charging EVs. Comparison of wireless charging systems across key metrics. IoT system accuracy and feedback delay In Fig. 14, the real-time temperature measured by the IoT-integrated sensor is compared to the actual measured temperature as charging was taking place for 10 min. Raising the workload steadily pushed the active temperature up from around 32 °C to about 52 °C. The IoT sensor’s temperature very closely aligned with the expected pattern, having a maximum deviation of 0.6 °C and an average error of 0.43 °C. By following the temperature closely, the DS18B20 sensor proves it is the ideal temperature sensor for WPT thermal control systems. Quick reaction by the IoT system plays a big role in avoiding sudden temperature changes. Because the fuzzy logic controller’s response time was very quick, it prevented the system temperature from rising above 55 °C. That is much better than the deviations up to ± 1.2 °C reported in their similar research on low-cost systems53. In addition, Fig. 15 shows each component’s share of the feedback latency which helps identify the main issues affecting response speed. The stage that caused the longest latency was MQTT transmission and reception with 28 ms, then came the actuator response delay of 24 ms and finally NodeMCU processing time at 20 ms. Getting readings from the voltage, current and temperature sensors was completed within 20 ms. The system was designed so that the usual feedback loop time of 116 ms is within the required limit for near real-time performance. The Table 4 below summarizes the statistical perspective of the key performance indicators obtained based on comparative Figs. (3–8 and 14). It provides the standard deviation, the mean efficiency, the maximum surface temperature, and the response time of the wired and the static wireless and the IoT-adaptive wireless charging modalities. These quantitative syntheses of canonical figures of merit have been widely used in the modern electric vehicle wireless power transfer studies to make the graphical results more interpretable. Actual versus IoT-sensed temperature over time. Average feedback latency per component (ms). Table 4Comparative summary of key performance metrics.ConfigurationMean efficiency (%)Std. Deviation (%)Max temperature (°C)Response time (ms)Wired charging89.4 ± 1.359.2152Static wireless90.2 ± 1.156.7140IoT–adaptive wireless (Proposed)93.5 ± 0.852.0116 Comparative summary of key performance metrics. Cost-efficiency trade-off The charge technology of electric-vehicles must be evaluated based on the efficiency and cost of installation. The trade-off of three configuration variants in Indian market versus global market is shown in Fig. 16 that compares the costs in the Indian market to that of the global market. Wired charging is the least expensive option, estimated to cost approximately 3500 dollars per system worldwide (in India it would approximately cost 2800 dollars) and it provides an efficiency of approximately 95%. Nevertheless, its hard-wired nature restricts movement and it can be unsafe in the event of wet ground or out the door. The immobile wireless systems have no cables in them, but must be installed with specific orientation and they have a set frequency; the cost of the immobile wireless systems is about 7800 worldwide with an efficiency of approximately 89%. The wireless system based on the IoT is more flexible and a superior trade-off: costing 9500 across the globe (8800 in India) and with a 93.5 efficiency, close to the wired charging. The increased price is associated with the addition of NodeMCU microcontrollers, sensors, MQTT modules, and real-time control. Beyond the costs of more than 12,000 are not associated with extra efficiency, which implies that the adaptive design proposal is costly with a high efficiency ratio. Cost (₹) versus efficiency (%) for static wireless and IoT adaptive. Conclusion The study presented the conceptualization, design, and optimization of an EV charging system, built on an IoT-wireless system. Experiments revealed that the IoT system had a better performance in different loads than other strategies with a power transfer efficiency of 93.5% which is much greater than the 89% of power transfer demonstrated in the case of the static wireless systems. They ensured the stability of the system, whereby the efficiency deviations remained within the range of ± 1.8 0–1.8 irrespective of the variation in the load. The RSM in the form of the Central Composite Design (CCD) showed that the coil gap, input voltage, and load level are the most important factors to explain more than 96 plicate efficiency variance. The adjusted R 2 of 0.948 is used to show that the model can be applied to the unobservable operating conditions. ANOVA provided the evidence that the same factors and interactions between them have a significant effect on the performance of the system. The ideal settings, as identified by 3D response surface and contour plots, were coil gap of 18 mm, 24 V and 75% load and actual performance was only 2.3% below the ideal settings, which reflects the accuracy of the model. Thermal control There was a notable improvement in the field of thermal control: the system kept the temperatures in the coils to less than 52 C with the help of an IoT-based control, ensuring thermal runaway. The error in temperature tracking was between −0.6 to + 0.6 o C and the load balancing and actuator adjustments time was 116 ms. The analysis of the cost showed that, though the implementation costs are higher, the IoT-based design would provide a great deal of energy efficiency, minimize charging problems, and provide a better control over the current. The present paper presents a prototype of an IoT-based adaptive wireless charger that can perform effectively and safely in the laboratory. Even though it is not a commercial product, the system has good potential to scale and lower costs in future electric-vehicle charging networks, provided it experiences massive field testing. Nevertheless, the testbed today is only functioning at sub-kilowatt scales and has been experimented in indoor controlled environments. To apply it practically, we have to test it with increased power, with different environmental conditions, and use it as a part of large-scale EV battery systems. Such steps have to do with ensuring the reliability in the long run and EMI. Using the real-time sensor metrics, the control algorithms, and statistical optimization, we show that smart, data-driven wireless charging can significantly improve the performance, flexibility and reliability in the next-generation mobility solutions. The charging capacity of the prototype is only limited to sub-kilowatt which is a disadvantage to its applicability in the charging of electric-vehicles. Its relevance could be enhanced by considering space-tested systems with a power of greater than 1 kW. External environmental variables are deliberately avoided in the study which is temperature, humidity, and altitude. These aspects control the stability of the outdoor system in relation to electromagnetic coupling and the heat dissipation that are critical in the operation of the system outdoors. Non-compliance is caused by dynamic electromagnetic coupling, charge loss without supervision, system motion without supervision and underutilization of capacity. This uncertainty is one of the major challenges to stable functioning in the framework of regulations. Acknowledgements The authors extend their appreciation to the Deanship of Research and Graduate Studies at King Khalid University for funding this work through Large Research Project under grant number RGP2/487/46. Abbreviations Internet of Things (–) Response surface methodology (–) Central composite design (–) State of charge (%) Power transfer efficiency (%) Voltage (V) Current (A) Coil surface temperature (°C) Input power (W) Output power (W) Analysis of variance (–) Author contributions Conceptualization, P.V.N., A.C.; Writing—Review and Editing, P.S.A., M.S.K.; Methodology, S.K.J., Y.D.J.; Formal analysis, M.G., T.M.Y.K.; Supervision, Y.D.J. Funding The authors extend their appreciation to the Deanship of Research and Graduate Studies at King Khalid University for funding this work through Large Research Project under grant number RGP2/487/46. Data availability The data that supports the findings of this study are available within the article. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The data that supports the findings of this study are available within the article.