Method for detecting a fault of a system

Abstract

A method for detecting and identifying a fault of a system comprising at least one sensor for more efficient maintenance.

Description

METHOD FOR DETECTING A FAULT OF A SYSTEMTechnical field

[0001] The invention lies in the field of system monitoring and, more specifically, it relates to a method for detecting and identifying a fault in a supervised system wherein sensors provide a continuous data output.

[0002] Sensors are used in a variety of applications where accurate and reliable data is essential for system performance, including but not limited to environmental monitoring, industrial systems, robotic systems, and automated control systems.Background of the invention

[0003] Sensors are fundamental components of modern data-driven technologies, serving as a crucial component in different applications, such as industrial manufacturing, automotive systems, aerospace, and environmental monitoring. These devices are essential for ensuring accurate data acquisition, system efficiency, and operational safety. However, sensors are susceptible to faults caused by a range of factors, including environmental conditions (such as extreme temperatures, humidity, or dust), physical wear and tear, hardware degradation over time, electrical issues (such as voltage fluctuations or electromagnetic interference), and even software or firmware malfunctions. Such faults can lead to serious consequences, including compromised system performance, extended machine downtime, safety risks, and diminished reliability [1], [2],

[0004] Sensor-related failures contribute significantly to operational inefficiencies in large industrial facilities. On average, these facilities could lose more than a day of production each month due to machine failures, including issues with sensors. This downtime leads to significant financial losses, with Fortune Global 500 manufacturing and industrial firms facing an estimated $864 billion annually equivalent to 8% of their total revenue [3],

[0005] To address these challenges, extensive research has been conducted on Fault Detection and Diagnosis, FDD, methods and fault tolerance techniques. FDD methods aim to identify faults, isolate faulty components, and classify the fault type and characteristics. Fault tolerance techniques complement these efforts by ensuring system performance and reliability are maintained despite anomalies.

[0006] FDD has evolved significantly, offering a variety of techniques to enhance the reliability of sensor systems. Known approaches may be efficient and adaptable for different applications, but they face challenges such as high computational complexity and limited generalizability across diverse applications in addition to the need of the data reconstruction.

[0007] A system’s reliability is commonly reinforced via hardware redundancy, which involves using multiple sensors to monitor the same parameter, ensuring reliability. For example, a study in [4] proposed a redundancy metric to design robust multi-sensor systems, although it faces challenges related to computational complexity. Despite its effectiveness, hardware redundancy increases cost, weight, and system complexity, making it less practical for resource-constrained environments like robotics and the Internet of Things, loT.

[0008] Alternative approaches such as analytical redundancy and observer-based methods use mathematical models to detect and mitigate sensor faults, thereby decreasing the dependence on physical redundancy. Observer-based fault reconstruction and estimation methods [1], [2] [6, 7] are particularly effective but often require precise system modelling and significant computational resources. Known hybrid approaches, which combine hardware and analytical redundancy, achieve high detection accuracy but remain costly and complex to implement [5], Advanced methods utilizing intelligent algorithms and large datasets are also emerging but demand substantial computational power, making them less practical for real-time applications.

[0009] Furthermore, many current methods are quite specialized and tailored for specific applications or systems. For example, aerospace applications prioritize fault modelling and detection to ensure mission-critical safety, lithium-ion batteries employ targeted methods to maintain energy efficiency, and automotive perception systems rely on advanced fault recovery techniques to enhance safety and reliability. Although these approaches are effective in their respective areas, their specialization restricts adaptability to other fields, highlighting the necessity for versatile solutions that can tackle a wide array of fault scenarios at limited complexity.

[0010] Despite significant advancements in sensor fault diagnosis and tolerance, several limitations persist in the current state of the art. These drawbacks affect the practical implementation, reliability, and are herein summarized.

[0011] Limited applicability: Many existing methods are tailored to specific systems or applications, making them less versatile or adaptable across diverse systems and sensor types.

[0012] High cost and physical complexity: Solutions relying on hardware redundancy, while effective, significantly increase the overall cost, weight, and physical complexity of the system. This is particularly disadvantageous for applications with strict constraints on these factors, such as robotics, aerospace, or portable systems.

[0013] High computational demands: Known advanced techniques, especially those employing signal-based methods or data-intensive intelligent algorithms, often demand substantial computational resources. This requirement makes real-time implementation impractical in many cases.

[0014] These limitations highlight the need for more versatile, efficient, and comprehensive fault-healing solutions that can be applied across various systems. An ideal approach should not only detect faults but also provide its identification.Technical problem to be solved

[0015] It is an objective to present a method, which overcomes at least some of the disadvantages of the prior art. In particular, it is an objective to provide a solution which ensures the integrity of continuous sensor readings by detecting faults that may arise during operation.Summary of the invention

[0016] In accordance with a first aspect of the invention, a method for detecting a fault of a system comprising at least one sensor is provided. The method comprises the steps: a) providing, at a diagnostic device, a sensor signal from the at least one sensor, wherein the sensor signal is raw data collected directly by the at least one sensor; b) providing, at the diagnostic device, a first reference signal; c) generating, using computing means, a difference signal between the sensor signal and the first reference signal; d) if the generated difference signal triggers a pre-defined fault threshold, i.e., being above or below a pre-defined threshold: outputting a fault alert; providing a correlated sensor signal; generating a reference signal by comparing the provided sensor signal with the correlated sensor signal, e.g. by calculating at least one statistical metric such as a Pearson Correlation Coefficient; if the generated reference signal triggers a pre-defined fault threshold: identifying the fault alert as a sensor fault; if thegenerated reference signal does not triggers a pre-defined fault threshold: identifying the fault alert as a system fault. This step provides important maintenance information because it directly identifies whether the fault is due to sensor failure or a malfunction in other parts of the system. If the generated difference signal does not triggers a pre-defined fault threshold: outputting the provided sensor signal.

[0017] Herein, the first reference signal is a signal from a physical model (preferential for systems with predictable / stable measurements); a data-driven model (preferential for systems with previous / historical data), e.g., a machine learning model-generated signal pre-trained on historical signal data of the sensor or a statistical model; or a redundant sensor signal (preferably for chaotic / unpredictable systems), i.e. a sensor signal measuring the same physical variable.

[0018] The physical model may be a mechanical model, a thermodynamic model, or electromagnetic model.

[0019] If the physical model is a mechanical model the correlated sensor may be a load cell, position sensor, velocity sensor, acceleration sensor, gyroscope, encoder, or a combination of these.

[0020] If the physical model is a thermodynamic model the correlated sensor may be a temperature sensor, pressure sensor, humidity sensor, or a combination of these.

[0021] If the physical model is an electromagnetic model the correlated sensor is magnetic field sensor, electric field sensor, or a combination of these.

[0022] In a further embodiment, the method for detecting a fault of a system further comprising as preceding offline steps: acquiring synchronized data from a plurality of sensors measuring different physical variable over a pre-defined period, preferably the sensors operate under identical or comparable conditions for meaningful correlation analysis; comparing two sensors from the plurality of sensors, each measuring a different physical variable, by calculating at least one statistical metric, e.g., Pearson Correlation Coefficient; outputting a pair of correlated sensors, corresponding to the compared two sensors, if the calculated statistical metrics triggers a pre-defined threshold, e.g., an absolute Pearson correlation coefficient exceeding 0.7. This set of steps establishes the relevant correlated sensor pair of the provided sensor signal with nofurther information on the system physics which is particularly important for unpredictable physical systems. This means that, when the method is running there is a known to be correlated sensor, from a different physical variable, that is used to identify a fault as a sensor or system fault.

[0023] The synchronized data may be normalized and / or pre-processed, e.g., denoising. This operation my be particularly relevant for signals that are on different scales / magnitudes.

[0024] In a second aspect of the invention, a system for detecting a fault comprising a diagnostic device for providing sensor signals, at least one sensor, and computing means adapted to execute the any of the steps of the method for detecting a fault of a system herein disclosed.

[0025] In an embodiment, the system for detecting a fault further comprising at least two correlated sensors, wherein each sensor is configured to measure a correlated physical quantity.

[0026] In a third aspect of the invention, a computer program is provided, comprising instructions which, when the program is executed cause a computer to carry out the steps of the method for detecting a fault of a system herein disclosed.

[0027] In a fourth aspect of the invention, a computer-readable storage medium is provided, comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method for detecting a fault of a system herein disclosed.

[0028] By using the proposed invention, it becomes possible to provide an efficiently and continuously monitor sensor data to detect a fault, wherein a fault corresponds to an anomaly. The process avoids overly complex algorithms, ensuring low computational requirements while maintaining robust performance.

[0029] The present method, after a detection of an anomaly, or fault, determines the fault’s origin, whether it is from the sensor itself or from another component in the supervised system that could influence the sensor data. This step is crucial to accurately diagnose the issue and initiate an healing process.

[0030] The disclosed method presents several key advantages over existing state-of-the-art solutions. Its versatility allows to adapt it to a wide range of sensor types and applications, including environmental monitoring and physically supervised systems, making it broadly applicable. One of its most significant features is rapid fault detection, with detection times of at least one-hundredth of a second in tested scenarios, i.e. detection in real-time. Additionally, the module ensures fast recovery, with recovery times ranging from 0 to 0.1 seconds, enabling realtime operation even under challenging fault conditions. Herein, recovery is to be understood as a signal which passes from being a faulty signal to an healthy signal.

[0031] Unlike traditional methods that rely only on hardware redundancy, the provided solution minimizes hardware dependency, reducing both costs and system complexity. Furthermore, its low computational requirements make it an efficient option for resource-constrained systems. These combined advantages establish the disclosed method as a practical, and highly effective solution for addressing sensor faults in modern applications.Brief description of the drawings

[0032] Several embodiments of the present invention are illustrated by way of figures, which do not limit the scope of the invention, wherein:figure 1 shows a flowchart of an embodiment of a method for detecting a fault of a supervised system;figure 2 shows a flowchart of an embodiment according to the invention comprising a further step of fault identification is performed;figure 3 shows a flowchart of an embodiment according to the invention comprising a further step of sensor data correction.Detailed description of the invention

[0033] This section describes aspects of the invention in further detail based on preferred embodiments and on the figures. The figures do not limit the scope of the invention. Throughout the description, like numerals will be used to describe like concepts in different embodiments. Details that are described in the context of a particular embodiment are applicable to other embodiments, unless otherwise stated.

[0034] The description puts focus on those aspects that are relevant for understanding the invention. It will be clear to the skilled person that a system for detecting a fault also comprises other commonly known aspects, such as connectors / ports, signal conditioning circuits, an appropriately dimensioned power supply, analog-to-digital converter, control ler / processor, or an output interface, even if those aspects are not explicitly mentioned.

[0035] Figure 1 represents a flowchart of an embodiment of a method for detecting a fault of a system, wherein 101 represents a faulty sensor signal, 102 represents a diagnostic device, and 103 represents a fault alert.

[0036] The physical sensor can be defined according to the environment where it is measuring, herein described as the operational environment. Thus, being categorized into two types: predictable environment sensors, or unpredictable environment sensor.

[0037] For the Predictable Environment Sensors, these sensors may be integrated into physical systems where their behaviour is well-defined and consistent. They are used in controlled or stable environments where operating conditions and system interactions are predictable.Examples include sensors used in industrial systems, robots, and other cyber-physical systems, CPS, where system dynamics are controlled and can be accurately modelled.

[0038] For the Unpredictable Environment Sensors, these sensors may be deployed in environments where conditions can unpredictably change and are more susceptible to external influences. Usually, they measure environmental parameters such as temperature, humidity, CO2 levels, or dust, where fluctuations in the environment introduce greater uncertainty. To address this, fault detection and identification processes are very important. In some cases, redundant sensors can be used to ensure accurate data even when one sensor fails.

[0039] In both environments, which will be later discussed in detail, the physical sensor measures at least one physical parameter, generating a Measured Sensor Data (200). This data may contain inaccuracies or faults due to various factors such as physical wear and tear, hardware degradation, electrical issues (e.g., voltage fluctuations or electromagnetic interference), calibration drift, or software and firmware malfunctions.

[0040] Advantageously, the present method not only detects but further identifies the fault, thus ensuring measurement reliability. Firstly, it detects a fault, i.e. , by continuously monitoringsensor data for anomalies, if no fault is detected, the system is in a healthy state. Secondly, if an anomaly is detected, the fault is identified whether it originates from the sensor itself or from the system, which may indicate external factors affecting the sensor’s output.

[0041] Figure 2 shows a flowchart according to the invention comprising a further step of fault identification. Herein, 200 represents sensor data acquired by a sensor, also named as received sensor signal, 201 represents a fault detection module, 202 represents an output comprising the provided sensor data, 203 represents a fault identification module, 204 represents an output identifying the fault alert as a system fault, and 205 represents an output identifying the detected fault as a sensor fault.

[0042] The fault detection module (201) may be implemented as a computational module comprising an electronic processor, memory, and associated software or firmware instructions configured to analyse sensor signals for anomalies.

[0043] The electronic processor executes algorithms that monitor the signal characteristics in real time, applying methods such as threshold-based detection, statistical anomaly detection, or machine learning models, as described herein.

[0044] The memory may store a reference signal (300), historical signal data, predefined fault patterns, and correction parameters used for comparison and decision-making. The fault detection module (201) and the fault identification module (203) may be integrated within a sensor system itself or operate as part of an external signal processing unit, communicating via wired or wireless interfaces. Upon detecting a fault and identifying the fault type, the fault detection module (201) and / or fault identification module (203) may log diagnostic data, and initiate corrective actions, such as signal filtering, recalibration, or redundant data substitution, to ensure continuous and reliable sensor operation.

[0045] The method begins with the provided sensor data (200), which is the raw data collected by a physical sensor. This data may contain inaccuracies or faults due to various factors. Here, the fault detection (201) module continuously monitors this provided sensor data to detect any anomalies. If no fault is detected, the system is considered to be unflawed, i.e. , healthy, and the output of this module is the same as the provided sensor data.

[0046] If no fault is detected, i.e., the generated difference signal does not trigger a pre-defined fault threshold, the system is in a healthy condition, and the output of the fault detection module (201) is the same as the provided sensor data.

[0047] If a fault is detected, the process moves to the next steps of fault identification (203) and further analysis. This module determines whether the fault lies with the sensor itself or not, e.g., in another related system component.

[0048] For both the fault detection (201) module and the fault identification (203) module the pre-defined fault threshold maybe a dynamic threshold, i.e., varying over time.

[0049] In another embodiment, a fixed constant threshold may be used via a residual signal, i.e., the difference between the second reference signal and the provided sensor signal, and then comparing this residual signal with the constant threshold to detect faults.

[0050] Even in the case of a stuck fault, e.g., a freeze sensor signal, a residual signal is calculated, i.e., a difference between the provided and a reference signal. If the stuck signal remains close to the reference, the residual stays below the threshold, and no fault is detected. However, if the reference signal changes while the stuck sensor remains constant, the residual increases, exceeding the threshold and triggering a fault detection.

[0051] Figure 3 shows a flowchart of an embodiment according to a further aspect of the invention comprising a further step of sensor data correction, wherein 300 represents a first reference signal, 401 represents a difference signal generation module, 402 represents an output of a healthy sensor data, 403 represents an input for a second reference signal, 404 represents reference signal generation module, 405 represents an alert for a faulty system, 406 represents an output for an faulty sensor, and 407 represents a sensor data correction module.

[0052] To detect the fault (201), a first reference signal (401) representing the expected healthy behaviour of the sensor is provided. This first reference signal may be generated through one of the following approaches:

[0053] Physics-Based Approach - Embodiment 1: If the system's dynamic behaviour is well understood, a mathematical model (e.g., an observer or state estimator) can be used to calculate the expected sensor output.

[0054] Data-Driven Approach - Embodiment 2: When historical data from fault-free operation is available, machine learning, or statistical models, can be trained to predict the expected sensor behaviour.

[0055] Redundant Sensors - Embodiment 3: In systems with multiple sensors measuring the same parameter, the data from redundant sensors can serve as the healthy reference signal.

[0056] As previously addressed, two main types of sensors are discussed: predictable environment sensors and unpredictable environment sensors. The use of each one of these will depend on their application and the system they operate in. These two sensors correspond to different embodiments which cater to different requirements for generating the healthy signal, i.e. , the reference signal, used in the fault identification module (203).

[0057] Embodiments 1 and 2

[0058] These embodiments have a superior performance for sensors that monitor and measure parameters in controlled and / or supervised systems such as robotics, industrial machinery, automotive systems, and other similar cyber-physical systems, CPS. In these cases, the generation of a first reference signal (401) can leverage a physical model (embodiment 1) or data-driven approaches (embodiment 2), e.g., by applying a machine learning model pre-trained on historical signal data of the sensor. These methods work effectively because the behaviour of supervised systems is well-defined and predictable and / or modellable.

[0059] The fault identification (203) module may classify a system fault if it does not trigger a threshold from a second reference signal, i.e., the fault is likely to be in the system being monitored. This may indicate a broader issue affecting multiple sensors. In an embodiment, the module may trigger a fault alarm warning to alert operators for external intervention. This ensures the broader system issue can be addressed promptly.

[0060] Alternatively, the fault identification (203) module may classify a sensor fault if it triggers a threshold from a second reference signal, i.e., the fault is isolated to the sensor itself, suggesting that it is malfunctioning.

[0061] In an embodiment, the method may proceed a fault estimation and correction (407). These steps correct the sensor’s output, allowing the system to continue functioning reliably without interruption.

[0062] Fault identification (203) is the method of determining whether an anomaly in the sensor data (200) originates from the sensor itself or from another part of the system being monitored. Fault identification (203) specify the source of the issue, enabling appropriate corrective action.

[0063] Embodiment 3

[0064] This embodiment has a superior performance for sensors used in environmental monitoring applications, such as temperature, humidity, or dust sensors. These sensors do not operate within a physically predictable system, so generating a healthy signal requires a more robust strategy. Here, at least one additional redundant sensor measuring the same physical parameter is required to act as a first reference signal. This method is practical for environmental sensors, as there are no physical models to leverage for predicting the expected behaviour of the sensor signal.

[0065] When a fault is detected, it means there is a discrepancy between the outputs of two physical sensors, in this case redundant sensors. The fault identification process is responsible for determining whether one sensor is faulty, and the other is still providing accurate data, or if both sensors are malfunctioning. This process is particularly relevant for possible later sensor signal corrections.

[0066] Embodiment 4

[0067] To identify the fault (203), e.g., if it is a faulty sensor or not, a second reference signal (403) representing the expected healthy behaviour of the sensor is input. This second reference signal is generated through a correlated sensor signal.

[0068] Correlated sensors are sensors whose measurements are interdependent, i.e. , a change in one sensor’s data reflects in others. Identifying these correlated sensors may be used for fault detection and / or identification. Nevertheless, before real-time operations, i.e. in offline, it is crucial to identify these correlated sensors, which will are used in generating a difference signal (401) during an online stage.

[0069] In predictable physical systems, examples of correlated sensors include:Position, Velocity, and Acceleration Sensors: In a mechanical system (e.g., automotive), these sensors are inherently connected. For instance, acceleration impacts velocity, which in turn influences position, as known from Newton’s law of movement.- Temperature and Pressure Sensors: In a thermal system (e.g., a boiler), temperature and pressure are correlated, as known from Boyle’s law.- Angular Velocity and Position Sensors: In robotic systems, gyroscopes (angular velocity) and encoders (angular position) are correlated. Changes in angular velocity directly affect angular position.

[0070] Examine the signal data from at least one correlated sensor, previously identified during an offline stage to determine whether at least one correlated sensor also shows an anomaly.

[0071] In unpredictable physical systems, examples of correlated sensors include:- Temperature and Humidity Sensors: In HVAC systems, temperature and humidity sensors are correlated because changes in temperature can affect humidity levels. - CO2 and Oxygen Sensors: In air quality monitoring, CO2 levels and oxygen levels are correlated, as certain activities that increase CO2 (like combustion) can decrease oxygen levels.

[0072] In an embodiment, after generating a second reference signal (404), it is generated a residual signal, also named as a second difference signal, by comparing the provided sensor data (200) to the second reference signal (404), for example, by subtracting the provided sensor data (200) to the second reference signal (404), i.e. , Residual Signal = Provided Sensor Data -Second Reference Signal.

[0073] This residual signal represents a deviation of the actual sensor data from its expected behaviour and it can be evaluated to determine if a fault on sensor is present by thresholding it with a predefined threshold.

[0074] The threshold may be defined as a deviation up to 10% from the reference data, preferably 5%, more preferably 3% and most preferably 1%. The deviation is chosen based on the sensor's datasheet, historical data, and system characteristics. When the sensor’s provided data exceeds this threshold, the system detects a fault.

[0075] In an embodiment the threshold has an upper limit different of a lower limit, i.e. it is not symmetrically triggered around the healthy signal or reference signal.

[0076] Alternatively, a Machine Learning, ML, model is used to identify the behaviour of the supervised system by learning the relationship between its inputs and outputs via training with historical sensor data.

[0077] For example, in the case of a simple motor, a ML model can be trained to represent the motor's behaviour. This model takes as input an electrical current and predicts an output, e.g., speed. Here, to detect faults in the speed sensor, an estimated output from the ML model is compared with the actual output from the real motor’s sensor. Since both receive the same input, the electrical current, any significant difference between the estimated and actual speed indicates a fault.

[0078] In an embodiment, the historical signal data is fed to the machine learning model as a sliding window of a pre-defined length.

[0079] After a sensor fault is detected via the provided sensor data, the following steps may be performed to identify the source of the fault:- Single Sensor Fault: If one redundant sensor passes the correlation test (i.e., it still verifies the nominal conditions), it is considered healthy, and the other sensor is identified as faulty.Both Sensors Faulty: If neither redundant sensor passes the correlation test, both are considered faulty. In this case, trigger a fault alarm warning to avoid using faulty data for further analysis and decision-making.

[0080] If the fault is from the system, the fault identification module (203) may trigger a fault alarm or warning for operators about potential system issues.

[0081] If the fault is from the sensor, the fault identification module (203) may send the faulty sensor signal for correction (407).

[0082] In a particular embodiment, the process of sensor fault detection starts by detecting an anomaly. Herein, a sensor, which measures a specific physical variable, is provided. The provided signal is compared with a first reference signal. If the residual signal exceeds a predefined threshold, a fault is detected. At this stage, it is not known if it’s a sensor fault or a system fault, so a further step of fault identification is applied.

[0083] For generating the first reference signal, there are three possible cases for obtaining a first reference signal:Predictable / Stable Systems: If the sensor is integrated into a system with stable behaviour, a physical model can be developed to estimate the reference signal.- Systems with Historical Data: If historical data is available, machine learning models can be used to predict the expected output.Redundant Sensors: For unpredictable environment sensors such as the environmental monitoring sensors, e.g., temperature, humidity, dust, at least one redundant sensor is used as a reference signal.

[0084] In cases of predictable / stable systems and systems with historical data, if there is a difference between the reference signal and the provided signal, it indicates an anomaly.However, it may be needed to determine whether the issue is with the sensor itself or if it originates within the system and affects the sensor’s output.

[0085] For identifying the fault type, at least one correlated sensor is used to identify whether the fault is in the sensor or the system:If a correlated sensor also shows anomalies, the issue is likely with the system.If the correlated sensor is not affected, the fault is in sensor itself.

[0086] For unpredictable environment sensors, where there is no accurate physical modelling of the system like in a predictable / stable system, or systems with historical data, a redundant sensor is used as a first reference. If the two sensors show different values, either one is faulty, or both are faulty. To confirm, a correlation analysis with other related sensors may be performed. If the sensor passes correlation checks, it is healthy; otherwise, it is faulty.

[0087] To validate the effectiveness of the present method comprehensive laboratory tests were performed on three different types of sensors, with data collected and analysed using MATLAB / Simulink R2024b, chosen for its advanced capabilities in modelling, simulation, and analysis.

[0088] In these examples, for each sensor type, a particular corresponding embodiment of the invention was applied. To thoroughly evaluate a self-healing strategy, i.e. , a method for detecting, identifying and correcting faulty sensor data, a comprehensive and challenging faultscenarios were designed for each sensor. These scenarios aimed to test the invention’s robustness and effectiveness in executing the fault detection under realistic conditions.

[0089] The primary outcome evaluated of the tests was the fault detection by identifying anomalies in sensor data in real time. The evaluation metrics used to quantify the performance was the detection time, i.e., the time taken to detect the fault. This metric demonstrate the method’s capability to maintain accurate and reliable sensor outputs even in the presence of faults.

[0090] Example F1 - Unpredictable environment sensors

[0091] For this example, the unpredictable environment sensor chosen was the temperature sensor Model STTS751. In this case, a threshold-based of X = 1% method was used.

[0092] A challenging fault scenario (F1) was designed to simulate real-world sensor issues, incorporating a series of faults occurring successively at different times and with varying durations. The faults also differ in type:Drift: A gradual change in sensor readings over time, which could be caused by wear and tear of the sensor. For instance, in temperature sensors, this can occur when the sensor's internal components degrade over time, leading to a slow, continuous shift in its measurements.Intermittent Bias: A sudden offset in sensor readings, often due to power issues or voltage fluctuations. For example, a temperature sensor might experience a bias when the sensor's power supply fluctuates, causing the sensor to report a constant deviation from the true value.Random Fault: This fault can arise from external noise or environmental factors like dust, electromagnetic interference. Such noise can cause irregular, unpredictable variations in the sensor’s data.

[0093] In this scenario, the detection time was of 0.01s. This demonstrates that the disclosed method is highly efficient in detecting for faults in an unpredictable environment such as a temperature sensor, ensuring that the system can continue to operate reliably, even in the presence of complex faults.

[0094] Example F2 - Unpredictable environment sensors

[0095] The fault detection threshold was set at ±1%, ensuring sensitivity to small deviations while evaluating the system’s capability under stringent conditions. When the provided signal exceeded this limit, the fault was immediately detected, and a possible correction could begin without delay.

[0096] To evaluate the effectiveness of the present correction method on humidity sensors, a challenging fault scenario F2 that combines three common fault types, occurring successively at different times and with varying durations was applied.Drift: A gradual deviation in sensor readings over time.Loss of Accuracy: This represents a reduction in the sensor’s ability to provide precise measurements, often due to sensor degradation or calibration issues after prolonged use.Freezing: The sensor output remains constant despite changes in the actual humidity level, which may be due to power supply interruptions or firmware / software malfunctions.

[0097] In this scenario, the detection time was of 0.01s.

[0098] Example F3 - Predictable environment sensors

[0099] In this fault scenario, two faults were considered: a drift, which represents gradual inaccuracies over time, which could result from sensor wear and tear or aging electronic; and a random fault, which simulates loss of accuracy due to external disturbances or electromagnetic interference.

[0100] Example F4 - Predictable environment sensors

[0101] In this fault scenario an intermittent bias was considered. Here, a sequence of sudden constant offsets were applied, simulating voltage fluctuations or connector issues in the sensor.

[0102] The detection threshold was set at ±5%, reflecting the operating specifications and characteristics of the joint position sensor. Fault detection occurred immediately once the sensor output exceeded this limit.

[0103] The performance of the present disclosure for the joint position sensor under fault conditions F3 and F4 was assessed using key metrics. Fault detection occurs immediately upon exceeding the predefined threshold, ensuring a rapid response, in this case of 0.1s. Recoverytime for both fault scenarios is measured at 0.1 seconds, demonstrating the system's ability to quickly correct faulty signals.

[0104] Throughout this document, the term "anomaly" is used interchangeably with "fault" and should be interpreted as a synonym in the context of this disclosure.

[0105] It should be noted that features described for a specific embodiment described herein may be combined with the features of other embodiments unless the contrary is explicitly mentioned. Based on the description and on the figures that have been provided, a person with ordinary skills in the art will be enabled to implement a computer program for executing the described methods without undue burden and without requiring additional inventive skill.

[0106] It should be understood that the detailed description of specific preferred embodiments is given by way of illustration only, since various changes and modifications within the scope of the invention will be apparent to the person skilled in the art. The scope of protection is defined by the following set of claims.

[0107] References[1] Fault Detection and Diagnosis Methods for Sensor Systems: A Scientific Literature Review (2023)[2] Sensor Fault Detection and Diagnosis: Methods and Challenges (2024)[3] World’s Largest Manufacturers Lose Almost $1 Trillion a Year to Machine Failures," Automation.com, Jun. 2021. [Online], Available: https: / / www.automation.com / enus / articles / june-2021 / world-largest-manufacturers-lose- almost-1 -trillion. Accessed: Nov. 11, 2024.[4] Sensor Fault Detection, Isolation, and Estimation in Lithium-Ion Batteries (2016)[5] A Redundancy Metric Set within Possibility Theory for Multi-Sensor Systems (2021) [6] A review of fault types and analytical and hardware redundancy for the sensor functional safety (2024)[7] Sensor Fault Reconstruction Using Robustly Adaptive Unknown-Input Observers (2024)

Claims

Claims1. Method for detecting a fault of a system comprising at least one sensor, comprising the steps:a) providing, at a diagnostic device, a sensor signal (200) from the at least one sensor; b) providing, at the diagnostic device, a first reference signal (300);c) generating, using computing means, a difference signal (401) between the sensor signal (200) and the first reference signal (300);d) if the generated difference signal triggers a pre-defined fault threshold:outputting a fault alert;providing a correlated sensor signal (403);generating a second reference signal (403) by comparing the provided sensor signal (200) with the correlated sensor signal (403);if the generated reference signal (404) triggers a pre-defined fault threshold:identifying the fault alert as a sensor fault (406);e) if not, identifying the fault alert as a system fault (405) if not, outputting the provided sensor signal (402);wherein the first reference signal (300) is a signal from a physical model, a data-driven model, or a redundant sensor signal.

2. Method according to the previous claim wherein the data-driven model is a machine learning model-generated signal pre-trained on historical signal data of the sensor.

3. Method according to the previous claim, further comprising as preceding steps:acquiring synchronized data from a plurality of sensors measuring different physical variable over a pre-defined period;comparing two sensors from the plurality of sensors, each measuring a different physical variable, by calculating at least one statistical metric;outputting a pair of correlated sensors, corresponding to the compared two sensors, if the calculated statistical metrics triggers a pre-defined threshold.

4. Method according to the previous claim, wherein the synchronized data is normalized and / or pre-processed.

5. Method according to any of the previous claims, wherein the second difference signal (404) is generated by subtracting the provided sensor signal (200) from the generated second reference signal (403).

6. Method according to any of the previous claims, wherein the physical model is a mechanical model, thermodynamic model, or electromagnetic model.

7. Method according to the previous claim, wherein when the physical model is a mechanical model, the correlated sensor is a load cell, position sensor, velocity sensor, acceleration sensor, gyroscope, encoder, or a combination of these.

8. Method according to claim 5, wherein when the physical model is a thermodynamic model, the correlated sensor is a temperature sensor, pressure sensor, humidity sensor, or a combination of these.

9. Method according to claim 5, wherein when the physical model is an electromagnetic model, the correlated sensor is magnetic field sensor, electric field sensor, or a combination of these.

10. System for detecting a fault comprising a diagnostic device for providing sensor signals, at least one sensor, and computing means adapted to execute the any of the steps of the method of claims 1-8.

11. System according to the previous claim comprising, at least two correlated sensors, wherein each sensor is configured to measure a correlated physical quantity.

12. Computer program comprising instructions which, when the program is executed cause a computer to carry out the steps of the method for fault detection of any of the claims 1- 8.

13. Computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method for fault detection of any of the claims 1-8.

Images