Predictive failure detection system for voltage-controlled switches

Abstract

A system for predicting remaining service life of voltage-controlled switches (32) comprises a data collection module (62) that continuously collects measurement data including voltage, current, temperature, and switching frequency during switch operation. A machine learning model trained on historical data predicts switch aging based on the collected measurement data. An evaluation module compares aging of multiple switches, while a utilization adjustment module adjusts switch utilization based on predicted aging to prevent premature failure. A data integration module continuously updates the machine learning model with new measurement data collected during operation, enabling adaptive learning and improved prediction accuracy over time. This predictive approach allows proactive switch management, extending overall system lifespan and reducing the risk of unexpected failures.

Description

M / PUIN-023-PC- 1 -PREDICTIVE FAILURE DETECTION SYSTEM FOR VOLTAGE-CONTROLLED SWITCHES5 FIELD OF INVENTION

[0001] The present disclosure relates to predictive failure detection for voltage- controlled switches, and more particularly to a system, method, and computer program product for detecting premature failure in voltage-controlled resistors using machine learning models and intelligent utilization adjustment.BACKGROUND

[0002] Voltage-controlled switches, such as field-effect transistors (FETs), are5 widely used in various electrical and electronic systems to control the flow of current. These switches play a crucial role in managing power distribution, circuit protection, and system functionality across diverse applications ranging from consumer electronics to industrial equipment and aerospace systems. As technology advances and systems become more complex, the reliability and longevity of these switches0 have become increasingly important.

[0003] In many modern applications, voltage-controlled switches are employed in redundant configurations to ensure system reliability. This approach allows for continued operation even if one or more switches fail, by switching to equivalent backup elements. However, this redundancy strategy alone does not address the5 underlying issue of switch failure prediction and prevention.

[0004] The current approach to managing switch failures typically involves emergency shutdown procedures or deliberate tripping of defective switches to prevent short circuits and avoid more significant damage. While this method can prevent catastrophic failures, it often results in the complete loss of functionality for0 the affected switch and the element it controls. This can be particularly problematic in situations where there is a sudden, short-term increase in demand, as the system may not have sufficient capacity to meet these peak requirements.M / PUIN-023-PC- 2 -

[0005] In some cases, the failure of a single switch can lead to the shutdown of the entire system, necessitating downtime for repairs. This approach can result in significant operational disruptions, increased maintenance costs, and potential safety risks, especially in critical applications such as aerospace systems where5 continuous operation is essential.

[0006] Furthermore, the lack of predictive capabilities in current switch management systems often leads to either premature replacement of components that still have useful life remaining or, conversely, the continued use of switches that are at high risk of imminent failure. This inefficiency in lifecycle management results in increased costs and potential reliability issues.

[0007] DE 11 2023 000 002 T5 discloses a method in which information about the wear of computer nodes is collected. Based on this information, the computer nodes are classified according to their respective expected remaining service life. The method further includes reconfiguring the system5 in response to an event that involves either adding a replacement node or removing an existing node, wherein the reconfiguration is performed based on the determined ranking of the computer nodes.

[0008] DE 10 2018 217 336 A1 discloses a method for calculating the remaining service life of a switch, in which lifetime-relevant events are0 recorded and evaluated with respect to their frequency of occurrence. Based on this evaluation, it is determined whether the likelihood of future occurrences of such events is taken into account or disregarded in the calculation of the remaining service life.

[0009] It has been appreciated that a predictive failure detection system for5 voltage-controlled switches is needed that overcomes one or more of these problems.SUMMARY

[0010] According to one embodiment, a system for predicting remaining service0 life of voltage-controlled switches comprises a data collection module configured to continuously collect measurement data including voltage, in particular drain source voltage when an FET is used as a voltage-controlled switch (voltage and currentM / PUIN-023-PC- 3 - allowing to calculate the inner resistance of the FET in the ON state), current, temperature, and switching frequency during operation of the voltage-controlled switches; a machine learning model trained on historical data and updated with ongoing measurements to predict switch aging based on the collected measurement5 data; an evaluation module configured to compare aging of multiple switches; a utilization adjustment module configured to adjust switch utilization based on predicted aging to prevent premature failure; and a data integration module configured to continuously update the machine learning model with new measurement data collected during operation of the voltage-controlled switches. This system advantageously enables proactive management of switch usage, extending the overall lifespan of the system and reducing the risk of unexpected failures.

[0011] In a preferred embodiment, the machine learning model is trained on historical data collected from a test chamber simulating various operating conditions5 for the voltage-controlled switches. This facilitates more accurate predictions by incorporating a wide range of potential operating scenarios, enhancing the model's robustness and reliability.

[0012] The system may further include a utilization adjustment module configured to calculate a switch utilization priority value between 0 and 1 for each0 switch based on the predicted aging, wherein the priority value indicates the preferential use of the switch a; and prioritize the use of switches with higher priority values over switches with lower priority values. This approach allows for intelligent load distribution, optimizing the use of switches based on their predicted remaining lifespan. 5

[0013] In another embodiment, switches with priority values below a preset limit are no longer used. This feature prevents the use of switches that are at high risk of failure, further enhancing system reliability.

[0014] The system may also include a data integration module configured to continuously update the machine learning model with new measurement data0 collected during operation of the voltage-controlled switches. This continuousM / PUIN-023-PC- 4 - learning process enables the model to adapt to changing conditions and improve its predictive accuracy over time.

[0015] In a further embodiment, the system comprises an ETL (Extract, Transform, Load) pipeline configured to extract measurement data from the data5 collection module; transform the extracted data by cleansing, formatting, and structuring it for analysis; and load the transformed data into the machine learning model for predicting switch aging. This pipeline ensures that the data used for predictions is consistently high-quality and properly formatted, improving the overall accuracy of the system.

[0016] The system may further include a test environment comprising climate chambers configured to simulate various operating conditions for the voltage- controlled switches. This feature enables comprehensive testing and data collection under controlled conditions, enhancing the training dataset for the machine learning model. 5

[0017] In another preferred embodiment, the system includes a test orchestration system configured to monitor, orchestrate, and allocate tests to test hardware. This component streamlines the testing process, ensuring efficient use of resources and comprehensive coverage of test scenarios.

[0018] The system may also comprise a database configured to store test data0 and operational data. This centralized data storage facilitates easy access to historical information, enabling trend analysis and long-term performance tracking.

[0019] In a further embodiment, the system includes an interface configured to make trained models available for use. This feature allows for easy deployment and updating of models across multiple systems or locations. 5

[0020] The system may additionally comprise a prediction component configured to accept models in live or offline operation, consume measurement data from a data source, execute the model to rate the switches, and forward the ratings to an execution component; a controller component configured to accept ratings for the switches, prioritize the use of switches according to the ratings, and forwardM / PUIN-023-PC- 5 - commands to the execution component; and an execution component configured to control switches, generate measurement data, link measurement data with execution data and switch identity, and forward data to the database. This comprehensive architecture enables seamless integration of predictive capabilities5 into the operational workflow of the system.

[0021] In another embodiment, the utilization adjustment module is further configured to allow short-term phases of high utilization for individual switches based on their predicted aging. This feature provides flexibility in managing peak load periods while still maintaining overall system health.

[0022] The system may include an evaluation module further configured to determine optimal timing for replacement of an entire circuit board or if feasible individual switches based on the predicted aging of multiple switches on the circuit board. This capability enables proactive maintenance scheduling, minimizing downtime and optimizing resource allocation. 5

[0023] In a preferred embodiment, the machine learning model is further configured to take into account the aging of components other than the voltage- controlled switches in predicting switch aging. This holistic approach provides a more comprehensive assessment of system health and potential failure points.

[0024] The system may collect measurement data that further includes input0 current, climate chamber temperature, voltage drop, and creep current. These additional data points enhance the accuracy and comprehensiveness of the aging predictions.

[0025] In another embodiment, the machine learning model comprises a Recurrent Neural Network (RNN) architecture with multi-step forecast capability.5 This advanced model architecture is particularly well-suited for time-series data and can provide more accurate long-term predictions.

[0026] The Recurrent Neural Network (RNN) architecture may comprise a Long Short-Term Memory (LSTM) model. LSTM models are particularly effective atM / PUIN-023-PC- 6 - capturing long-term dependencies in time-series data, further improving prediction accuracy.

[0027] The system may further comprise an alert generation module configured to generate an alert when the number of available switches falls below a5 predetermined threshold. This feature ensures timely intervention when the system's operational capacity is at risk.

[0028] In a preferred embodiment, the system includes a risk threshold definition module configured to define measurement values indicating too high a risk for continued operation of the switches. This module adds an additional layer of safety by establishing clear criteria for switch deactivation.

[0029] The utilization adjustment module may be further configured to calculate a switch utilization priority value for each switch based on the predicted aging; prioritize the use of switches with higher priority values during periods of high demand; and intelligently distribute load among available switches to maximize5 overall system lifespan. This sophisticated load management strategy optimizes system performance while extending its operational life.

[0030] In some aspects, the voltage-controlled switches may be field-effect transistors (FETs), although other types of voltage-controlled switches may also be used. 0

[0031] According to another embodiment, a method for predictive failure detection of voltage-controlled switches is provided. The method comprises continuously collecting measurement data including voltage, current, temperature, and switching frequency during operation of the voltage-controlled switches; predicting switch aging using a machine learning model trained on historical data;5 evaluating and comparing aging of multiple switches; calculating for each switch a switch utilization priority value between 0 and 1 indicating the preferential use of the switch; and adjusting switch utilization based on the calculated priority values to prevent premature failure. This method enables proactive management of switch usage, extending system lifespan and reducing unexpected failures.M / PUIN-023-PC- 7 -

[0032] The further method and computer program product embodiments given in the dependent claims provide similar advantages as outlined above, offering flexible implementation options for various operational contexts and technological environments.5

[0033] Further details and advantages of the invention are become apparent from the following purely exemplary and non-limiting description of embodiments in conjunction with the drawing comprising six figures, which comprises 4 figures.BRIEF DESCRIPTION OF THE DRAWING

[0034] FIG. 1 illustrates a flowchart of a process for predicting and managing remaining service life of voltage-controlled switches, according to aspects of the present disclosure.

[0035] FIG. 2 depicts in more detail, but still quite schematically, a system diagram for simulating, monitoring, and predicting the remaining service life of5 voltage-controlled switches according to aspects of the present disclosure.

[0036] FIG. 3 illustrates a system diagram for predicting and managing aging of voltage-controlled switches, according to aspects of the present disclosure.DETAILED DESCRIPTION 0

[0037] The present disclosure provides a system, method, and computer program product for predictive failure detection of voltage-controlled switches. This system leverages machine learning models and intelligent utilization adjustment to predict the remaining service life of voltage-controlled switches. The system continuously collects measurement data, including voltage, current, temperature,5 and switching frequency, during the operation of the switches. This data is then used by a machine learning model, trained on historical data, to predict the aging of the switches.

[0038] An evaluation module compares the aging of multiple switches, and a utilization adjustment module adjusts the switch utilization based on the predictedM / PUIN-023-PC- 8 - aging to prevent premature failure. This proactive management of switch usage extends the overall lifespan of the system and reduces the risk of unexpected failures.

[0039] In some embodiments, the machine learning model is trained on5 historical data collected from a test chamber simulating various operating conditions for the switches. This approach enhances the model's robustness and reliability by incorporating a wide range of potential operating scenarios.

[0040] In other embodiments, the system includes a utilization adjustment module that calculates for each switch a switch utilization priority value based on the predicted aging. This priority value indicates the preferential use of the switch, and the system prioritizes the use of switches with higher priority values over switches with lower priority values. This intelligent load distribution optimizes the use of switches based on their predicted remaining lifespan.

[0041] The system may also include a data integration module that 5 continuously updates the machine learning model with new measurement data collected during operation of the switches. This continuous learning process enables the model to adapt to changing conditions and improve its predictive accuracy over time.

[0042] The Recurrent Neural Network (RNN) may be implemented as a Long0 Short-Term Memory (LSTM) model specifically optimized for processing time-series data. The LSTM architecture comprises input layers for the continuous measurement data, hidden layers with forget gates for selective information deletion, input gates for controlling new information, and output gates for output control, as well as output layers for predicting switch aging. 5

[0043] The system may continuously calculate the drain-source resistance (Rds(on)) as a quotient of the measured drain-source voltage and drain current (Rds(on) = Vds / ld). Changes in Rds(on) over time indicate aging effects of the FET switch and serve as a key parameter in the aging prediction model.M / PUIN-023-PC- 9 -

[0044] The ETL (Extract, Transform, Load) pipeline may comprise: (1) extraction of raw measurement data from sensors, (2) transformation through data cleansing, normalization, and feature engineering, and (3) loading of structured data into the machine learning model. The pipeline processes data streams in real-time5 and ensures data quality through outlier detection and plausibility checks.

[0045] The system, method, and computer program product disclosed herein provide a solution to the problem of premature failure of voltage-controlled switches, enhancing system reliability and longevity. The following sections provide a more detailed description of the operation and benefits of this predictive failure detection system.

[0046] Referring to FIG. 1, the flowchart illustrates a process for predicting and managing the remaining service life of voltage-controlled switches. The process begins at step S10 with the production of data in a controlled environment, such as a test chamber. This initial data is generated under various operating conditions to5 simulate the real-world usage of the switches. The data may include measurements of voltage, temperature, switching frequency, and other relevant parameters.

[0047] The collected data is then processed at in a "Collect I Clean / Transform Data" step S12. In this step, the raw data is cleaned, formatted, and transformed into a suitable format for further analysis. This may involve removing outliers,0 normalizing values, and converting data into a format that can be readily processed by a machine learning model.

[0048] Following the data preparation step, the process moves to a "Train Model" step S14. In this step, a machine learning model, such as a Recurrent Neural Network (RNN) architecture, is trained on the prepared data. The model5 learns to recognize patterns and relationships in the data that can be used to predict the aging of the switches. The training process may involve adjusting the model's parameters to minimize the difference between the model's predictions and the actual observed data.

[0049] The process then transitions to a "Collect Data while operating" step0 S16. In this step, data is continuously collected during the operation of the switches.M / PUIN-023-PC- 10 -This real-time data collection allows the system to monitor the switches' performance and condition in real-time, providing up-to-date information for the predictive model.

[0050] The collected operational data is then used in a "Use Model to Decide5 Further Actions" step S18. In this step, the trained model analyzes the operational data and makes predictions about the remaining service life of each switch. These predictions are used to make decisions about how to manage the switches to prevent premature failure.

[0051] Following the decision-making step, the process moves to a "Rate each switch" step S20. In this step, each switch is evaluated and assigned a rating based on its predicted remaining service life. The rating may be a numerical value or a categorical label indicating the preferential use of the switch.

[0052] The final step in the process is the "Use the switches according to rating" step S22. In this step, the utilization of the switches is adjusted based on5 their individual ratings. Switches with higher ratings may be prioritized for use, while switches with lower ratings may be used less frequently or not at all. This intelligent adjustment of switch utilization helps to maximize the overall lifespan of the system and prevent premature failures.

[0053] The process depicted in FIG. 1 is cyclical, with the "Use the switches0 according to rating" step S22 feeding back into the "Collect Data while operating" step S16. This feedback loop allows for continuous monitoring and adjustment of switch usage based on ongoing data collection and analysis. In some aspects, the process may also include steps for updating the machine learning model with new data, recalculating switch ratings, and adjusting the utilization of switches in5 response to changes in their predicted aging.

[0054] Referring to FIG. 2, the system diagram illustrates a test chamber setup for simulating and monitoring the usage of voltage-controlled switches. The test chamber setup includes a number of climate chambers 30, each including various components for data collection and analysis, such as a voltage controlled switch, in0 particular an FET switch 32, a voltage meter 34, a switch temperature sensor (in theM / PUIN-023-PC- 11 - shown embodiment a thermistor) 36, a switch input current source 38, a load 40, and a climate chamber temperature sensor (in the shown embodiment also a thermistor) 42.

[0055] In some aspects, the voltage meter 34 measures the voltage across the5 switch 32, providing valuable data on the electrical characteristics of the switch 32 during operation. The temperature sensors 36 and 42 monitor the temperature of the switch 32 respectively the temperature in the chamber 30, which can be indicative of the thermal load on the switch 32. A current sensor (not shown) measures the current flowing through the circuit, providing additional data on the electrical load on the switch 30 and data regarding the switching performance showing signs of ageing.

[0056] FIG. 2 also illustrates steps of using the system for predicting remaining service life of voltage-controlled switches. A climate control interface (not shown) regulates the environmental conditions within the chamber 30, allowing a step S445 of simulating the usage of switches various operating in various conditions. This can include varying the temperature, humidity, and other environmental factors within the chamber to simulate different real-world operating conditions for the switches.

[0057] The exact timing of voltage input and voltage size at the switch, along with the precise timing of measurement data is monitored (step S46). 0

[0058] The system simulates the usage of switches in the test chambers until all switches reach states in which they would no longer be operable (step S48). This process provides valuable data on the lifespan and failure modes of the switches under various operating conditions. If this process takes too long, the system may also use older switches or new ones artificially aged before testing. 5

[0059] The system collects all data generated during the simulation and monitoring phases (step S50). This collected data includes measurements of voltage, temperature, current, and other relevant parameters. In some aspects, the measurement data further includes input current and climate chamber temperature. This comprehensive data collection allows for a detailed analysis of switch behavior0 and lifespan under various operating conditions.M / PUIN-023-PC- 12 -

[0060] The collected data is converted in step S52 into a format readable by the model. This ensures that the data from various sensors and measurements, including voltage, temperature, and switching frequency, can be properly processed by the subsequent components. In some aspects, the climate control interface may5 also handle data related to voltage drop and creep current, providing additional insights into the electrical characteristics of the switches.

[0061] In step S54 an RNN architecture is used. The respective RNN architecture represents the core of the machine learning model. It utilizes a structure such as an LSTM (Long Short-Term Memory) model with a multi-step forecast capability, aimed at predicting distant developments in switch performance. The architecture is designed to accommodate multiple switch types if necessary, enhancing the versatility of the system.

[0062] A model training process S56 is responsible for training the RNN model using the formatted data from step S52. This process allows the model to learn5 patterns and relationships in the historical data to make accurate predictions. The model training process S56 may involve iterative learning and refinement based on new data, enabling the model to adapt to changing conditions and improve its predictive accuracy over time.

[0063] In a risk threshold definition step S58 measurement values are defined0 as thresholds that would indicate too high a risk for continued operation of the switches. These thresholds are used in step S60.

[0064] S60 is s step of providing an algorithm that uses the forecast form the machine learning model to determine, how long it will take a switch to reach the thresholds, and thus to determine when a switch may be approaching failure or5 inefficient operation. In some cases, the risk threshold definition S58 may be dynamically adjusted based on the predicted aging of the switches, allowing for more precise and responsive switch management.

[0065] The forecast algorithm provided in step S60 utilizes the predictions from the trained machine learning model to determine how long it will take for a switch to0 reach the defined risk threshold values. This algorithm provides insights into theM / PUIN-023-PC- 13 - remaining service life of each switch, enabling proactive switch management to prevent premature failure.

[0066] In operation, the system continuously collects and processes data through the climate control interface, feeding it into the RNN architecture, which has5 been trained through the model training process. The forecast algorithm then uses the model's predictions in conjunction with the risk threshold definition to assess the future performance and potential failure of the voltage-controlled switches. This integrated approach to data collection, processing, and analysis enables the system to predict switch aging with high accuracy and adjust switch utilization accordingly to maximize system lifespan and reliability.

[0067] The forecast algorithm is a key component of the system. This algorithm utilizes the predictions from the trained machine learning model to determine how long it will take for a switch to reach the defined risk threshold values. This prediction provides insights into the remaining service life of each switch, enabling5 proactive switch management to prevent premature failure. In some aspects, the forecast algorithm may also take into account other factors, such as the aging of other components, to provide a more comprehensive prediction of switch performance.

[0068] The system may be operable in both live operation with real-time data0 processing and offline operation with historical data. In live operation, the system processes measurement data in real-time and provides immediate predictions and control decisions. In offline operation, the system analyzes historical data to refine models and generate comprehensive aging assessments.

[0069] The machine learning model may comprise a multi-step forecast5 capability for predicting switch aging over multiple time horizons. This enables both short-term operational decisions and long-term maintenance planning by providing predictions ranging from minutes to months ahead.

[0070] The system may generate alerts when the number of operational switches falls below a safety threshold for continued system operation. These alertsM / PUIN-023-PC- 14 - enable proactive intervention before system capacity becomes critically compromised.

[0071] In some cases, the system may be implemented as a computer program product. The computer program product comprises a computer readable storage5 medium having program instructions embodied therewith. These program instructions are executable by a processor to cause the processor to perform the method of predictive failure detection of voltage-controlled switches as described herein.

[0072] The program instructions may include instructions for continuously collecting measurement data, predicting switch aging using a machine learning model, evaluating and comparing aging of multiple switches, calculating for each switch a switch utilization priority value indicating the preferential use of the switch, and adjusting switch utilization based on the calculated priority values to prevent premature failure. 5

[0073] In some aspects, the program instructions may also include instructions for continuously updating the machine learning model with new measurement data collected during operation of the voltage-controlled switches, defining risk threshold values indicating too high a risk for continued operation of the switches, and generating an alert when the number of available switches falls below a 0 predetermined threshold.

[0074] In this way, the system can be implemented as a computer program product, providing a flexible and scalable solution for predictive failure detection of voltage-controlled switches. This implementation allows the system to be easily integrated into existing infrastructure, enhancing system reliability and longevity5 without requiring significant hardware modifications.

[0075] Referring to FIG. 3, the system diagram illustrates a predictive failure detection system for voltage-controlled switches. The system comprises several interconnected components that work together to monitor, analyze, and control switch usage based on predicted aging.M / PUIN-023-PC- 15 -

[0076] The system includes a data collection module 62 configured to continuously collect measurement data during the operation of the switches 32. This module collects data from various sensors, including voltage meters 34, temperature sensors 36, and current sensors (not shown). The collected data includes voltage5 across each switch 32, temperature of the switches 32, and current flowing through the switches. This real-time data collection allows the system to monitor the switches' performance and condition in real-time, providing up-to-date information for the predictive model.

[0077] The collected data is then processed by an ETL (Extract, Transform, Load) process module 64. This module receives the collected data and prepares it for further analysis by cleansing, formatting, and structuring it for use by the machine learning model.

[0078] The prepared data is then fed into a forecast aging module 66, which predicts the aging of the switches based on the collected and processed data. This5 prediction is used by a calculate threshold module 68 to determine a switch utilization priority value for each switch. The switch utilization priority value, ranging from 0 to 1 , indicates for each switch the preferential use of the switch. In some aspects, a lower priority value may indicate a less preferential use of the switch, while a higher priority value may indicate that the switch should be used more0 preferred.

[0079] An adjust priority value module 70 takes the calculated priority values and adjusts them for each individual switch. These adjusted priority values are then used by a switch usage module 72 to determine how and when to use each switch based on its assigned priority value. In some aspects, switches with priority values5 below a preset limit are no longer used, effectively removing them from the pool of available switches 3. An alert may be generated when the number of available switches falls below a predetermined threshold. This intelligent adjustment of switch utilization helps to maximize the overall lifespan of the system and reduce the risk of unexpected failures. 0

[0080] The system incorporates a feedback loop, where the switch usage data is fed back into the data collection module 62 and the ETL process module 64,M / PUIN-023-PC- 16 - allowing for continuous monitoring and adjustment of the system. This feedback loop enables the system to adapt to changing conditions and improve its predictive accuracy over time.

[0081] In some cases, the system may also include a recommendation module5 that recommends replacement of switches based on their predicted aging and calculated threshold values. This proactive management of switch usage extends the overall lifespan of the system and reduces the risk of unexpected failures.

[0082] The system also includes an RNN (Recurrent Neural Network) architecture, which forms the core of the machine learning model. The RNN architecture is trained on historical data collected from a test chamber, allowing it to learn patterns and relationships in the data that can be used to predict the remaining service life of the switches.

[0083] In this way, the system provides a comprehensive solution for predictive failure detection of voltage-controlled switches, combining real-time data collection,5 advanced machine learning techniques, and intelligent utilization adjustment to optimize switch usage and prevent premature failures.

[0084] The climate chambers are integrated into the continuous learning process, generating accelerated aging data under controlled conditions including variable temperature, humidity, and switching cycles. This data continuously0 expands the training dataset of the machine learning model, enabling more accurate predictions across diverse operating conditions.

[0085] The risk assessment module defines threshold values that indicate too high a risk for continued operation of the switches. These thresholds are based on critical parameters such as maximum allowable Rds(on) values, temperature limits,5 and switching frequency constraints. When a switch approaches these threshold values, it is automatically flagged for reduced utilization or removal from service.

[0086] The test orchestration system provides automated management of testing procedures, monitoring test progress, allocating resources to test hardware, and ensuring comprehensive coverage of test scenarios. This system enablesM / PUIN-023-PC- 17 - efficient utilization of test equipment and systematic data collection across multiple test conditions.

[0087] The system also includes two power sources, in the shown example two 4.2 V power sources, connected to the switches 32. These power sources provide5 the operating voltage for the switches 32, enabling them to perform their switching functions. In some aspects, the power sources may be adjustable to simulate different operating conditions for the switches 32.

[0088] The prediction component may operate in both live and offline modes, accepting trained models and consuming measurement data from various sources. In live mode, it processes real-time data streams, while in offline mode, it analyzes historical datasets. The component executes the machine learning model to generate switch ratings and forwards these ratings to downstream components for decision-making.

[0089] The controller component may receive switch ratings and implements5 intelligent prioritization algorithms. During periods of high demand, it prioritizes switches with higher ratings while ensuring load distribution that maximizes overall system lifespan. The controller generates commands that are forwarded to the execution component for implementation.

[0090] The execution component may directly control the switches, generates0 measurement data during operation, and maintains linkage between measurement data, execution commands, and switch identity. This comprehensive data linkage enables precise tracking of switch performance and facilitates continuous model improvement.

[0091] The alert generation module may continuously monitor system status5 and generates warnings when the number of available switches approaches critical levels. These alerts include severity levels, estimated time until critical capacity is reached, and recommended actions for maintaining system reliability.

[0092] The interface component may provide standardized access to trained models, enabling deployment across multiple systems and locations. ThisM / PUIN-023-PC- 18 - component handles model versioning, compatibility checks, and secure model distribution to ensure consistent performance across different operational environments.

[0093] The order of the steps of the methods described herein is exemplary, but5 the steps may be carried out in any suitable order, or simultaneously where appropriate. Additionally, steps may be added or substituted in, or individual steps may be deleted from any of the methods without departing from the scope of the subject matter described herein.

[0094] Aspects of any of the examples described above may be combined with aspects of any of the other examples described to form further examples without losing the effect sought. Particular aspects of the invention can be listed and grouped as follows:

[0095] Aspect 1: A system for predicting remaining service life of voltage- controlled switches, comprising: 5 a data collection module configured to continuously collect measurement data including voltage, current, temperature, and switching frequency during operation of the voltage-controlled switches; a machine learning model trained on historical data to predict switch aging based on the collected measurement data; 0 an evaluation module configured to compare aging of multiple switches; and a utilization adjustment module configured to adjust switch utilization based on predicted aging to prevent premature failure. 5

[0096] Aspect 2: The system of aspect 1, wherein the machine learning model is trained on historical data collected from a test chamber simulating various operating conditions for the voltage-controlled switches.

[0097] Aspect 3: The system of aspect 1 or 2, wherein the utilization adjustment0 module is configured to:M / PUIN-023-PC- 19 - calculate a switch utilization priority value between 0 and 1 for each switch based on the predicted aging, wherein the switch utilization priority value indicates the preferential use of the switch; and prioritize the use of switches with higher priority values over switches5 with lower priority values.

[0098] Aspect 4: The system of aspect 3, wherein switches with priority values below a preset limit are no longer used. 0

[0099] Aspect 5: The system of any of aspects 1 to 4, further comprising a data integration module configured to continuously update the machine learning model with new measurement data collected during operation of the voltage-controlled switches.

[0100] Aspect 6: The system of any of aspects 1 to 5, further comprising an ETL (Extract, Transform, Load) pipeline configured to: extract measurement data from the data collection module; transform the extracted data by cleansing, formatting, and structuring it for analysis; and 0 load the transformed data into the machine learning model for predicting switch aging.

[0101] Aspect 7: The system of any of aspects 1 to 6, further comprising a test environment comprising climate chambers configured to simulate various operating5 conditions for the voltage-controlled switches.

[0102] Aspect 8: The system of any of aspects 1 to 7, further comprising a test orchestration system configured to monitor, orchestrate, and allocate tests to test hardware.

[0103] Aspect 9: The system of any of aspects 1 to 8, further comprising a database configured to store test data and operational data.

[0104] Aspect 10: The system of any of aspects 1 to 9, further comprising an5 interface configured to make trained models available for use.M / PUIN-023-PC- 20 -

[0105] Aspect 11: The system of any of aspects 1 to 10, further comprising: a prediction component configured to accept models in live or offline operation, consume measurement data from a data source, execute the model to5 rate the switches, and forward the ratings to an execution component; a controller component configured to accept ratings for the switches, prioritize the use of switches according to the ratings, and forward commands to the execution component; and an execution component configured to control switches, generate measurement data, link measurement data with execution data and switch identity, and forward data to the database.

[0106] Aspect 12: The system of any of aspects 1 to 11, wherein the utilization adjustment module is further configured to allow short-term phases of high utilization5 for individual switches based on their predicted aging.

[0107] Aspect 13: The system of any of aspects 1 to 12, wherein the evaluation module is further configured to determine optimal timing for circuit board replacement based on the predicted aging of multiple switches on the circuit board. 0

[0108] Aspect 14: The system of any of aspects 1 to 13, wherein the machine learning model is further configured to take into account the aging of components other than the voltage-controlled switches in predicting switch aging. 5

[0109] Aspect 15: The system of any of aspects 1 to 14, wherein the measurement data further includes input current, climate chamber temperature, voltage drop, and creep current.

[0110] Aspect 16: The system of any of aspects 1 to 15, wherein the machine0 learning model comprises a Recurrent Neural Network (RNN) architecture with multi-step forecast capability.

[0111] Aspect 17: The system of aspect 16, wherein the Recurrent Neural Network (RNN) architecture comprises a Long Short-Term Memory (LSTM) model.M / PUIN-023-PC- 21 -

[0112] Aspect 18: The system of any of aspects 1 to 17, further comprising an alert generation module configured to generate an alert when the number of available switches falls below a predetermined threshold.5

[0113] Aspect 19: The system of any of aspects 1 to 18, further comprising a risk threshold definition module configured to define measurement values indicating too high a risk for continued operation of the switches.

[0114] Aspect 20: The system of any of aspects 1 to 19, wherein the utilization0 adjustment module is further configured to: calculate a switch utilization priority value for each switch based on the predicted aging; prioritize the use of switches with higher switch utilization priority values during periods of high demand; and intelligently distribute load among available switches to maximize overall system lifespan.

[0115] Aspect 21 : The system of one of aspects 1 to 20, wherein the voltage- controlled switches are field-effect transistors (FETs). 0

[0116] Aspect 22: A method for predictive failure detection of voltage-controlled switches, comprising: continuously collecting measurement data including voltage, temperature, and switching frequency during operation of the voltage-controlled5 switches; predicting switch aging using a machine learning model trained on historical data; evaluating and comparing aging of multiple switches; calculating for each switch a switch utilization priority value between 0 and 1 indicating the preferential use of the switch; and adjusting switch utilization based on the calculated priority values to prevent premature failure.

[0117] Aspect 23: The method of aspect 22, wherein the machine learning model is trained on historical data collected from a test chamber simulating various5 operating conditions for the voltage-controlled switches.M / PUIN-023-PC- 22 -

[0118] Aspect 24: The method of aspect 22 or 23, further comprising continuously updating the machine learning model with new measurement data collected during operation of the voltage-controlled switches.5

[0119] Aspect 25: The method of any of aspects 22 to 24, wherein adjusting switch utilization comprises prioritizing the use of switches with higher priority values over switches with lower priority values.

[0120] Aspect 26: The method of aspect 25, wherein switches with priority values below a preset limit are no longer used.

[0121] Aspect 27: The method of any of aspects 22 to 26, further comprising generating an alert when the number of available switches falls below a 5 predetermined threshold.

[0122] Aspect 28: The method of any of aspects 22 to 27, wherein the measurement data further includes input current and climate chamber temperature. 0

[0123] Aspect 29: The method of any of aspects 22 to 28, wherein predicting switch aging comprises evaluating at least two of voltage drop, conducted current and creep current.

[0124] Aspect 30: The method of any of aspects 22 to 29, further comprising5 recommending replacement of switches based on their predicted aging and calculated threshold / priority values.

[0125] Aspect 31: The method of any of aspects 22 to 29, further comprising allowing short-term phases of high utilization for individual switches based on their predicted aging.

[0126] Aspect 32: The method of any of aspects 22 to 31, further comprising determining optimal timing for circuit board replacement based on the predicted aging of multiple switches on the circuit board. 5M / PUIN-023-PC- 23 -

[0127] Aspect 33: The method of any of aspects 22 to 32, further comprising taking into account the aging of components other than the voltage-controlled switches in predicting switch aging.5

[0128] Aspect 34: The method of any of aspects 22 to 33, further comprising: extracting measurement data from the data collection module; transforming the extracted data by cleansing, formatting, and structuring it for analysis; and loading the transformed data into the machine learning model for0 predicting switch aging.

[0129] Aspect 35: The method of any of aspects 22 to 33, further comprising: defining risk threshold values indicating too high a risk for continued operation of the switches; calculating a switch utilization priority value for each switch based on the predicted aging; prioritizing the use of switches with higher priority values during periods of high demand; and intelligently distributing load among available switches to maximize0 overall system lifespan.

[0130] Aspect 36: A method for implementing a predictive failure detection system for voltage-controlled switches, comprising: commissioning a machine learning model for a specific cell type; 5 preparing a user system to provide necessary measurement data for the machine learning model; making available software containing the machine learning model and a rating algorithm; and integrating the rating from the machine learning model into a decisionmaking process for controlling the switches.

[0131] Aspect 37: The method of aspect 36, further comprising: specifying cell types for testing; testing the specified cell types; and 5 integrating the predictive failure detection system into a user setup.M / PUIN-023-PC- 24 -

[0132] Aspect 38: A computer program product for predictive failure detection of voltage-controlled switches, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the5 program instructions executable by a processor to cause the processor to perform the method of any of aspects 22 to 38.

[0133] Aspect 39: A battery-powered land, sea, or air vehicle using a system according to any one of aspects 1 to 21 for predicting remaining service life of0 voltage-controlled switches and / or a method according to any one of aspects 22 to 35 for predictive failure detection of voltage-controlled switches.

[0134] It will be understood that the above description of a preferred embodiment is given by way of example only and that various modifications may be made by those skilled in the art. What has been described above includes examples of one or more embodiments. It is, of course, not possible to describe every conceivable modification and alteration of the above devices or methods for purposes of describing the aforementioned aspects, but one of ordinary skill in the art can recognize that many further modifications and permutations of various0 aspects are possible. Accordingly, the described aspects are intended to embrace all such alterations, modifications, and variations that fall within the scope of the appended claims

Claims

M / PUIN-023-PC- 25 -CLAIMS1. A system for predicting remaining service life of voltage-controlled switches, comprising:5 a data collection module configured to continuously collect measurement data including voltage, current, temperature, and switching frequency during operation of the voltage-controlled switches; a machine learning model trained on historical data to predict switch aging based on the collected measurement data; an evaluation module configured to compare aging of multiple switches; a utilization adjustment module configured to adjust switch utilization based on predicted aging to prevent premature failure; and a data integration module configured to continuously update the machine learning model with new measurement data collected during operation of the5 voltage-controlled switches.

2. The system of claim 1, wherein the utilization adjustment module is configured to: calculate a switch utilization priority value between 0 and 1 for each switch0 based on the predicted aging, wherein the switch utilization priority value indicates the preferential use of the switch; and prioritize the use of switches with higher priority values over switches with lower priority values. 5 3. The system of any of claims 1 to 2, further comprising an ETL (Extract, Transform, Load) pipeline configured to: extract measurement data from the data collection module; transform the extracted data by cleansing, formatting, and structuring it for analysis; and 0 load the transformed data into the machine learning model for predicting switch aging.

4. The system of any of claims 1 to 3, wherein the machine learning model comprises a Recurrent Neural Network (RNN) architecture with Long Short-TermM / PUIN-023-PC- 26 -Memory (LSTM) including input layers for the measurement data, hidden layers with forget gates, input gates and output gates, and output layers for aging prediction.

5. The system of any of claims 1 to 4, wherein the system continuously5 calculates drain-source resistance (Rds(on)) as a quotient of measured voltage and current and uses changes in Rds(on) as an aging indicator.

6. The system of any of claims 1 to 5, further comprising climate chambers for accelerated aging simulation that continuously generate additional training data for the machine learning model.

7. The system of any of claims 1 to 6, wherein the system is operable in both live operation with real-time data processing and offline operation with historical data.

58. The system of any of claims 1 to 7, further comprising a risk assessment module that defines threshold values indicating too high a risk for continued operation of the switches. 0 9. The system of any of claims 1 to 8, further comprising at least one of: a test orchestration system configured to monitor, orchestrate, and allocate tests to test hardware, a database configured to store test data and operational data, an interface configured to make trained models available for use, 5 an alert generation module configured to generate an alert when the number of available switches falls below a predetermined threshold.

10. The system of any of claims 1 to 9, further comprising: a prediction component configured to accept models in live or offline0 operation, consume measurement data from a data source, execute the model to rate the switches, and forward the ratings to an execution component; a controller component configured to accept ratings for the switches, prioritize the use of switches according to the ratings, and forward commands to the execution component; andM / PUIN-023-PC- 27 - an execution component configured to control switches, generate measurement data, link measurement data with execution data and switch identity, and forward data to the database.5 11 . The system of any of claims 1 to 10, wherein: the machine learning model is trained on historical data collected from a test chamber simulating various operating conditions for the voltage-controlled switches, and / or switches with priority values below a preset limit are no longer used, and / or the utilization adjustment module is further configured to allow short-term phases of high utilization for individual switches based on their predicted aging, and / or the evaluation module is further configured to determine optimal timing for circuit board replacement based on the predicted aging of multiple switches on the5 circuit board, and / or the machine learning model is further configured to take into account the aging of components other than the voltage-controlled switches in predicting switch aging, and / or the measurement data further includes input current, climate chamber0 temperature, voltage drop, and creep current, and / or the voltage-controlled switches are field-effect transistors (FETs).

12. The system of claim 2 or any of claims 3 to 11 when dependent on claim 2, wherein the utilization adjustment module is further configured to use the5 switch utilization priority values to: prioritize the use of switches with higher priority values during periods of high demand; and intelligently distribute load among available switches to maximize overall system lifespan.

13. A method for predictive failure detection of voltage-controlled switches, comprising: continuously collecting measurement data including voltage, current, temperature, and switching frequency during operation of the voltage-controlled5 switches;M / PUIN-023-PC- 28 - predicting switch aging using a machine learning model trained on historical data; evaluating and comparing aging of multiple switches; calculating for each switch a switch utilization priority value between 0 and 15 indicating the preferential use of the switch; adjusting switch utilization based on the calculated priority values to prevent premature failure; and continuously updating the machine learning model with new measurement data collected during operation of the voltage-controlled switches.

14. The method of claim 13, wherein the machine learning model comprises a Recurrent Neural Network (RNN) architecture with Long Short-Term Memory (LSTM). 5 15. The method of claim 13 or 14, further comprising continuously calculating drain-source resistance (Rds(on)) as a quotient of measured voltage and current.

16. The method of any of claims 13 to 15, further comprising generating additional training data through climate chambers for accelerated aging simulation.

017. The method of any of claims 13 to 16, wherein the method is performed in both live operation with real-time data processing and offline operation with historical data. 5 18. The method of any of claims 13 to 17, further comprising defining risk threshold values indicating too high a risk for continued operation of the switches.

19. The method of any of claims 13 to 18, wherein: the machine learning model is trained on historical data collected from a test chamber simulating various operating conditions, and / or adjusting switch utilization comprises prioritizing switches with higher priority values, and / or switches with priority values below a preset limit are no longer used, and / or the measurement data further includes input current and climate chamber5 temperature, and / orM / PUIN-023-PC- 29 - predicting switch aging comprises evaluating at least two of voltage drop, conducted current and creep current.

20. The method of any of claims 13 to 19, further comprising at least one of:5 recommending replacement of switches based on their predicted aging and calculated threshold / priority values, allowing short-term phases of high utilization for individual switches based on their predicted aging, determining optimal timing for circuit board replacement based on the predicted aging of multiple switches on the circuit board, taking into account the aging of components other than the voltage- controlled switches in predicting switch aging.

21. The method of any of claims 13 to 20, further comprising: 5 extracting measurement data from the data collection module; transforming the extracted data by cleansing, formatting, and structuring it for analysis; and loading the transformed data into the machine learning model for predicting switch aging.

022. The method of any of claims 13 to 21 , wherein the switch utilization priority values are used to: prioritize switches with higher priority values during periods of high demand; and 5 intelligently distribute load among available switches to maximize overall system lifespan.

23. The method of any of claims 13 to 22, wherein the machine learning model comprises multi-step forecast capability for predicting switch aging over multiple time horizons.

24. The method of any of claims 13 to 23, further comprising generating alerts when the number of operational switches falls below a safety threshold for continued system operation.M / PUIN-023-PC- 30 -25. A method for implementing a predictive failure detection system for voltage- controlled switches, comprising: commissioning a machine learning model for a specific cell type; preparing a user system to provide necessary measurement data for the5 machine learning model; making available software containing the machine learning model and a rating algorithm, wherein the machine learning model is continuously updated with new measurement data collected during the operation of the voltage-controlled switches; and 0 integrating the rating from the machine learning model into a decisionmaking process for controlling the switches.

26. A computer program product for predictive failure detection of voltage- controlled switches, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to perform the method of any of claims 13 to 24.

27. A battery-powered land, sea, or air vehicle using a system according to any0 one of claims 1 to 12 for predicting remaining service life of voltage-controlled switches and / or a method according to any one of claims 13 to 24 for predictive failure detection of voltage-controlled switches.

Images