Fault detection and diagnosis for the engine electrical system of a space launcher based on a temporal convolutional autoencoder and calibrated classifiers
Luis Basora, Louison Bocquet-Nouaille, Elinirina Robinson, Serge Le Gonidec · 2025
Read Fault detection and diagnosis for the engine electrical system of a space launcher based on a temporal convolutional autoencoder and calibrated classifiers by Luis Basora, Louison Bocquet-Nouaille, Elinirina Robinson, Serge Le Gonidec (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-13022.txt.
Introduction
Luis Basora, Louison Bocquet-Nouaille, Elinirina Robinson, and Serge Le Gonidec detect faults in a space launcher engine in two thousand twenty-five. The target is the electrical system that drives engine valves. Reusable launchers need onboard health checks of that system. Their design must report confidence, spot odd data, and curb false alarms.
A temporal convolutional autoencoder squeezes raw sensor data into traits. Two calibrated classifiers then flag and name faults from those traits. A conformal step marks out of range cases. A cumulative sum chart trims false alarms over time. Tests on simulated runs look promising, with real data still to come.
Catalog record
| CREATOR | Luis Basora, Louison Bocquet-Nouaille, Elinirina Robinson, Serge Le Gonidec |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 82 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-13022
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Fault detection and diagnosis for the engine electrical system of a space launcher based on a temporal convolutional autoencoder and calibrated classifiers Luis Basora, Louison Bocquet-Nouaille, Elinirina Robinson, Serge Le Gonidec Abstract In the context of the health monitoring for the next generation of reusable space launchers, we outline a first step toward developing an onboard fault detection and diagnostic capability for the electrical system that controls the engine valves. Unlike existing approaches in the literature, our solution is designed to meet a broader range of key requirements. This includes estimating confidence levels for predictions, detecting out-of-distribution (OOD) cases, and controlling false alarms. The proposed solution is based on a temporal convolutional autoencoder to automatically extract low-dimensional features from raw sensor data. Fault detection and diagnosis are respectively carried out using a binary and a multiclass classifier trained on the autoencoder latent and residual spaces. The classifiers are histogram-based gradient boosting models calibrated to output probabilities that can be interpreted as confidence levels. A relatively simple technique, based on inductive conformal anomaly detection, is used to identify OOD data. We leverage other simple yet effective techniques, such as cumulative sum control chart (CUSUM) to limit the false alarms, and threshold moving to address class imbalance in fault detection. The proposed framework is highly configurable and has been evaluated on simulated data, covering both nominal and anomalous operational scenarios. The results indicate that our solution is a promising first step, though testing with real data will be necessary to ensure that it achieves the required maturity level for operational use. 1 Introduction This research is part of a European effort to develop a health monitoring system (HMS) for the next generation of expandable and reusable space launchers. Currently, European launchers lack advanced onboard control or diagnostic systems, relying solely on post-flight data analysis through telemetry feeds. This limitation stems from the expendable nature of current launchers and the restricted telemetry bandwidth. However, an HMS will be essential for future reusable launchers and their engines to ensure high levels of system reliability and safety, as well as to reduce operational costs. The future HMS will need to cover both in-flight and post-flight phases, similar to existing practices in aeronautics. An overview of the usage context of the HMS in space propulsion systems is provided in Férard et al. 2021 . As part of the onboard HMS under development, this paper introduces a data-driven approach for the fault detection and diagnosis (FDD) of the electrical valve actuators in the engine of a space launcher. This approach is designed to operate effectively during both the transient and steady-flight phases, ensuring continuous monitoring and early detection of potential issues throughout different operational stages. Electrical valve actuators are critical components that regulate fuel flow and other essential fluids for engine operations. These actuators typically function with precision control under high-stress conditions, given the demanding and often extreme environments involved in launch systems. They operate as part of the engine’s electrical system, which must ensure accurate and reliable performance to maintain stability and prevent potential mission failures. Given these requirements, any faults in these actuators, such as malfunctions in their electromechanical components or issues in the control signals, could impact engine performance and safety significantly. FDD for valve actuators often relies on analyzing time series data from electrical and mechanical signals to identify any irregularities that may signal wear or failure. Within the framework of the European Space Agency’s (ESA) Future Launcher Preparatory Program (FLPP), several solutions were investigated to advance the development of FDD, including for rocket engine valve actuators. While machine learning algorithms, such as neural networks, were explored, most early efforts focused on model-based approaches ( Marcos et al. 2013 ) . More recently, artificial intelligence (AI) and machine learning (ML) methods have emerged as valuable tools in prognostics and health management (PHM) systems, bringing advanced predictive capabilities to fault detection. However, their implementation in critical systems, like those used in space launchers, remains challenging. Key barriers include ensuring data reliability, handling data limitations in fault scenarios, and maintaining interpretability and transparency in model outputs. Additionally, the integration of AI and ML solutions in complex applications demands rigorous testing and verification to ensure they can operate robustly under real-world conditions and meet stringent safety standards. Furthermore, ML algorithms like the one proposed in this paper should be designed to operate on an onboard electronic control unit (ECU). The proposed modeling approach is implemented as a flexible framework that enables comprehensive experimentation to optimize FDD performance. This framework allows for systematic tuning of selected models, ensuring they are optimally configured to the characteristics of the provided data. In the configuration phase, we run multiple experiments to guide model selection and adjust hyperparameters effectively. The evaluated models and techniques include: 1. Temporal convolutional autoencoder (TCAE) to extract informative low-dimensional features from raw sensor data representing electrical, mechanical and valve commands signals. 2. Histogram-based boosting trees (HGBT) classifiers for binary (fault detection) and multiclass (diagnosis) classification of health conditions. 3. Threshold moving to address class imbalance in fault detection in fault detection by simply adjusting the threshold. 4. Conformal anomaly detection (CAD) as a relatively simple and practical solution for OOD detection. 5. CUSUM (Cumulative Sum Control) to reduce false positives in fault and OOD detection, enhancing reliability. 6. Classifier calibration to output probabilities interpretable as confidence levels for predictions. To the best of our knowledge, combining TCAE with HGBT classifiers in FDD has not been explored before. Additionally, the application of CUSUM to the failure probability estimated by a binary classifier is a novel approach. Our study provides valuable insights into using techniques in the FDD field that have seen limited adoption, such as model calibration and conformal anomaly detection. In our study, we assess their benefits and limitations in enhancing the reliability of predictions and improving the overall fault detection and diagnosis capabilities. The initial evaluation is conducted using simulated data from a physical model of an electrically controlled engine valve. This serves as a preliminary step to assess the suitability of the proposed model pipeline. In the next phase, the approach will undergo further validation using data from an integrated test bench facility with electric motors that operate real valves. In this context, the flexibility of our framework is a key advantage, allowing it to be customized to fit the specific characteristics of different datasets. The paper is structured as follows. Section 2 reviews existing research on time series modeling and FDD, with a particular focus on electric motor applications using electrical signals such as currents and voltages. Section 3 outlines the key references and theoretical concepts behind the main techniques used in our framework, especially those that are less widely known. Section 4 presents the system under study and details the simulated data, including an analysis of class overlapping. In Section 5 , we describe our proposed modeling approach, highlighting the main steps of the inference process. Section 6 details the experiments conducted for model selection and hyperparameter optimization to configure the pipeline effectively. Section 7 evaluates the performance of the configured framework using the simulated dataset. Section 8 summarizes the key findings, discusses limitations, and outlines potential directions for future research. 2