Fault detection method and device, electronic equipment, storage medium and computer program product

Abstract

The invention discloses a fault detection method and device, electronic equipment, a storage medium and a computer program product, and belongs to the field of artificial intelligence. The fault detection method comprises the following steps: acquiring a to-be-detected signal of target equipment; determining a reconstruction error corresponding to the to-be-detected signal based on a signal obtained by reconstructing the to-be-detected signal; based on a reference reconstruction error, performing outlier detection on the reconstruction error to obtain an outlier detection result corresponding to the reconstruction error; the reference reconstruction error is determined based on a signal obtained by reconstructing a normal signal corresponding to the to-be-detected signal; and determining a fault detection result corresponding to the target equipment based on an outlier detection result corresponding to the reconstruction error. According to the invention, dependence on fault data can be avoided, and meanwhile, the accuracy, generalization and robustness of fault detection are improved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to a fault detection method, device, electronic device, storage medium and computer program product. Background Technology

[0002] The safe, stable, and efficient operation of industrial equipment is crucial for ensuring production continuity, and timely and accurate fault detection is the core technological link in achieving this goal. Currently, most fault detection methods for industrial equipment employ classification models for fault identification. The core of these methods lies in training the model with a large amount of labeled fault sample data, enabling it to classify real-time monitoring signals into "normal" or "faulty" states.

[0003] However, in real-world industrial applications, industrial equipment is typically in a long-term stable operating state, and fault sample data is scarce, making it difficult to meet the needs of model training, resulting in low accuracy of classification models. Summary of the Invention

[0004] This application aims to address at least one of the technical problems existing in the related art. To this end, this application proposes a fault detection method, apparatus, electronic device, storage medium, and computer program product, which can avoid dependence on fault data and improve the accuracy of fault detection.

[0005] Firstly, this application provides a fault detection method, the method comprising:

[0006] Acquire the signal to be tested from the target device;

[0007] Based on the signal reconstructed from the signal under test, determine the reconstruction error corresponding to the signal under test;

[0008] Based on the reference reconstruction error, outlier detection is performed on the reconstruction error to obtain the outlier detection result corresponding to the reconstruction error; the reference reconstruction error is determined based on the signal reconstructed from the normal signal corresponding to the signal under test.

[0009] Based on the outlier detection results corresponding to the reconstruction error, the fault detection results corresponding to the target device are determined.

[0010] According to the fault detection method of this application, determining the reconstruction error corresponding to the signal under test based on the signal reconstructed from the signal under test includes:

[0011] Based on the operating frequency of the target device, the signal to be tested is bandpass filtered to obtain a filtered signal.

[0012] The filtered signal is fused with the signal to be tested to obtain a fused signal;

[0013] The reconstruction error is determined based on the signal obtained by reconstructing the fused signal.

[0014] According to the fault detection method of this application, determining the reconstruction error based on the signal obtained from the fused signal reconstruction includes:

[0015] The fused signal is subjected to sliding window processing to obtain multiple signal segments of the fused signal;

[0016] Reconstruction is performed based on each of the signal segments to obtain the reconstructed signal segments corresponding to each of the signal segments;

[0017] Based on each of the reconstructed signal segments and the corresponding signal segments, the reconstructed error includes each reconstructed error element.

[0018] According to the fault detection method of this application, the step of reconstructing based on each of the signal segments to obtain the reconstructed signal segments corresponding to each of the signal segments includes:

[0019] Each of the signal segments is input into the reconstruction model to obtain each of the reconstructed signal segments output by the reconstruction model; the number of convolutional modules included in the reconstruction model is determined based on the signal-to-noise ratio of the signal to be tested, and the reconstruction model is trained based on the sample signal segments corresponding to the sample normal signal.

[0020] According to the fault detection method of this application, determining the fault detection result corresponding to the target device based on the outlier detection result corresponding to the reconstruction error includes:

[0021] If the proportion of outlier detection results in the outlier detection results corresponding to each of the reconstruction error elements exceeds the target proportion, the fault detection result corresponding to the target device is determined to be a fault.

[0022] According to the fault detection method of this application, the step of performing outlier detection on the reconstruction error based on the reference reconstruction error to obtain the outlier detection result corresponding to the reconstruction error includes:

[0023] The median and the corresponding absolute deviation of the median are determined from multiple reference error elements; the normal signal includes multiple normal signal segments, and the reference reconstruction error includes the reference error elements corresponding to the multiple normal signal segments;

[0024] Based on the difference between the reconstruction error and the median, and the absolute deviation of the median, the outlier value corresponding to the reconstruction error is determined;

[0025] Based on the comparison between the outlier value corresponding to the reconstruction error and the target threshold, the outlier detection result corresponding to the reconstruction error is determined.

[0026] Secondly, this application provides a fault detection device, which includes:

[0027] The signal acquisition module is used to acquire the signal to be tested from the target device.

[0028] An error determination module is used to determine the reconstruction error corresponding to the signal under test based on the signal reconstructed from the signal under test.

[0029] An outlier detection module is used to detect outliers in the reconstruction error based on a reference reconstruction error, and obtain an outlier detection result corresponding to the reconstruction error; the reference reconstruction error is determined based on the signal reconstructed from the normal signal corresponding to the signal under test.

[0030] The fault detection module is used to determine the fault detection result corresponding to the target device based on the outlier detection result corresponding to the reconstruction error.

[0031] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the fault detection method as described in the first aspect above.

[0032] Fourthly, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the fault detection method as described in the first aspect above.

[0033] Fifthly, this application provides a chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the fault detection method as described in the first aspect.

[0034] In a sixth aspect, this application provides a computer program product, including a computer program that, when executed by a processor, implements the fault detection method as described in the first aspect above.

[0035] The above-described one or more technical solutions in the embodiments of this application have at least one of the following technical effects:

[0036] Furthermore, by reconstructing the signal under test, the reconstruction error is calculated. Then, an outlier detection algorithm from statistics is introduced to determine the outlier nature of the reconstruction error relative to the reference reconstruction error calculated from the normal signal, and a standard for judging anomalies is given. This avoids dependence on fault data and improves the accuracy, generalization and robustness of fault detection.

[0037] Furthermore, the reconstruction algorithm and outlier detection algorithm are flexibly configurable and easily expandable, and can be widely adapted to fault detection and early warning of different industrial objects and signal types.

[0038] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0039] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0040] Figure 1 This is a flowchart illustrating the fault detection method provided in an embodiment of this application;

[0041] Figure 2 This is a schematic diagram of the structure of the reconstruction model provided in the embodiments of this application;

[0042] Figure 3 This is a comparison diagram of the time-domain waveforms before and after filtering provided in the embodiments of this application;

[0043] Figure 4 This is a schematic diagram of the structure of the fault detection device provided in the embodiments of this application;

[0044] Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0045] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0046] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0047] The fault detection method, fault detection device, electronic device, and readable storage medium provided in this application will be described in detail below with reference to the accompanying drawings and through specific embodiments and application scenarios.

[0048] The fault detection method can be applied to the terminal, and can be executed by the hardware or software in the terminal.

[0049] The fault detection method provided in this application embodiment can be executed by an electronic device or a functional module or entity in an electronic device that can implement the fault detection method. The electronic devices mentioned in this application embodiment include, but are not limited to, mobile phones, tablets, computers, cameras, and wearable devices. The fault detection method provided in this application embodiment is described below using an electronic device as the execution subject.

[0050] Currently, most fault detection methods for industrial equipment rely on training classification models using fault data to determine whether the measured signal belongs to the normal or faulty category; or they use single-classification methods to model normal data and determine whether the measured signal is abnormal. These methods have the following drawbacks:

[0051] 1. Training a classification model requires collecting a large number of fault samples or labeled data, but in actual industrial scenarios, there are often only normal samples and fault samples are scarce, making it difficult to meet the training requirements of the model.

[0052] 2. Single classification methods are limited by the model's capabilities when applied. For example, they cannot handle high-dimensional samples, have few adjustable parameters, and are difficult to adapt to different devices and working conditions.

[0053] 3. The methods used in the testing process are fixed, lacking a flexible modular structure and scalability.

[0054] To address the issues of existing industrial fault detection methods being overly reliant on fault data, having insufficient model generalization ability, and lacking flexibility and scalability, this application provides a fault detection method.

[0055] Figure 1 This is a flowchart illustrating the fault detection method provided in an embodiment of this application, as shown below. Figure 1 As shown, the fault detection method includes steps 110, 120, 130 and 140.

[0056] Step 110: Obtain the signal to be tested from the target device.

[0057] In practice, the target equipment, i.e., the equipment requiring fault detection, can be rotating machinery such as motors, bearings, and engines, or other types of industrial equipment such as compressors, centrifugal pumps, and CNC machine tools. This application does not specifically limit this. During the operation of the target equipment, electronic devices can collect vibration signals from the target equipment using signal acquisition devices such as vibration sensors or vibration sensor arrays to obtain the signal to be measured.

[0058] The signal to be tested can be obtained from the original acquisition by the signal acquisition device, or it can be obtained after preprocessing such as denoising on the signal acquired by the signal acquisition device. This application does not specifically limit this.

[0059] Step 120: Determine the reconstruction error corresponding to the signal under test based on the signal reconstructed from the signal under test;

[0060] In practice, the signal to be tested can be directly reconstructed, and the difference between the reconstructed signal and the signal to be tested can be used to determine the reconstruction error of the signal to be tested. Alternatively, the signal to be tested can be filtered, and the difference between the reconstructed signal and the filtered signal can be used to determine the reconstruction error. Or, the signal to be tested and the filtered signal can be fused to obtain a fused signal, and the difference between the reconstructed signal and the fused signal can be used to determine the reconstruction error. This application does not specifically limit the implementation of these methods.

[0061] The signal reconstruction method can be to use generative models such as autoencoders, variational autoencoders or generative adversarial networks to reconstruct the input signal, or iterative algorithms or other signal reconstruction methods. This application does not specifically limit the method in this regard.

[0062] Step 130: Based on the reference reconstruction error, outlier detection is performed on the reconstruction error to obtain the outlier detection result corresponding to the reconstruction error; the reference reconstruction error is determined based on the signal reconstructed from the normal signal corresponding to the signal under test.

[0063] Step 140: Based on the outlier detection results corresponding to the reconstruction error, determine the fault detection results corresponding to the target device.

[0064] In practice, outlier detection methods can be modified Z-scores, interquartile range-based methods, etc., and this application does not specifically limit them.

[0065] Specifically, considering that normal signal samples are relatively easier to obtain, the embodiments of this application can determine the reconstruction error based on the difference between the signal reconstructed from the normal signal and the normal signal, and then use this as a benchmark, i.e., the reference reconstruction error, and then apply the outlier detection method in statistics to determine the outlierness of the reconstruction error corresponding to the signal under test, thereby obtaining the outlier detection result used to characterize the outlierness of the reconstruction error, and finally determine whether the target device has a fault based on the outlier detection result, thereby obtaining the fault detection result corresponding to the target device.

[0066] It is understandable that by using the reference reconstruction error as a benchmark to judge the outlier of the reconstruction error corresponding to the signal under test, the stronger the outlier, the greater the deviation between the signal under test and the normal signal, and the more likely the target device is to be in a faulty state.

[0067] It should be noted that although the reconstruction error corresponding to the signal under test can reflect the degree of deviation between the signal under test and the normal signal to a certain extent, if the reconstruction error is directly used for fault judgment based on a fixed threshold, there is a defect that the threshold calibration depends on a large number of samples, and the generalization ability across devices and operating conditions is poor. Even if the source of the signal changes, the threshold needs to be adjusted. The embodiments of this application introduce an outlier detection method, which only requires a small number of normal signal samples to establish a judgment benchmark, and can achieve fault detection with robustness and interpretability across scenarios. Even if the signal source changes or the operating conditions deviate, a high fault detection rate can be maintained without adjustment. Even for hidden operating conditions where the equipment has not yet failed but there is a potential fault, the embodiments of this application can accurately determine the fault initiation state while the equipment is still running, and achieve predictive maintenance with zero downtime.

[0068] According to the fault detection method provided in the embodiments of this application, the signal under test is reconstructed, and the reconstruction error is calculated. Then, an outlier detection algorithm in statistics is introduced to determine the outlierness of the reconstruction error relative to the reference reconstruction error calculated from the normal signal, and a standard for judging anomalies is given. This avoids dependence on fault data and improves the accuracy, generalization and robustness of fault detection. In addition, the reconstruction algorithm and the outlier detection algorithm are flexibly configurable and easily expandable, and can be widely adapted to fault detection and early warning of different industrial objects and signal types.

[0069] In some embodiments, step 120 may include:

[0070] Based on the operating frequency of the target device, the signal to be tested is bandpass filtered to obtain the filtered signal.

[0071] The filtered signal and the signal under test are fused to obtain the fused signal;

[0072] The reconstruction error is determined based on the signal obtained from the fused signal reconstruction.

[0073] Specifically, a narrowband bandpass filter can be used to filter the signal to be tested, retaining only the components in the frequency band centered at frequency f and with width as the radius, thus obtaining a filtered signal, where f is the rotational speed frequency of the target device. Then, the filtered signal is fused with the signal to be tested to obtain a fused signal. Finally, the reconstruction error corresponding to the signal to be tested is determined based on the difference between the reconstructed signal and the fused signal.

[0074] In practice, if the signal to be measured comes from only one sensor, it can be directly bandpass filtered. If the signal to be measured comes from multiple sensors, the signal from each sensor can be used as a channel of the signal to be measured. For each channel, bandpass filtering is used to process it to obtain the filtered signal.

[0075] Signal fusion can be achieved by directly splicing two signals, doubling the number of channels, or by weighted averaging the two signals. This application does not specifically limit the method. After fusion, signal reconstruction can be performed directly based on the fused signal, and the reconstruction error of the signal under test can be determined by using the difference between the reconstructed signal and the fused signal. Alternatively, the fused signal can be processed by sliding window slicing or similar methods, and the reconstruction error can be determined by using the difference between the reconstructed signal and the processed signal. This application also does not specifically limit the method.

[0076] It should be noted that the filtered signal obtained based on the operating frequency of the target equipment will highlight the fault more clearly, reducing interference from useless frequency bands. Compared with the test signal alone, combining the test signal and the filtered signal for fault detection can achieve a higher fault detection accuracy.

[0077] In some embodiments, determining the reconstruction error based on the signal obtained from the fused signal reconstruction may include:

[0078] The fused signal is processed by sliding window to obtain multiple signal segments of the fused signal;

[0079] Reconstruct the signal segments based on each signal segment to obtain the reconstructed signal segments corresponding to each signal segment.

[0080] Based on each reconstructed signal segment and its corresponding signal segment, the various reconstructed error elements included in the reconstruction error are determined.

[0081] Specifically, to improve the accuracy of fault detection, the reconstruction error of the signal under test can be determined as follows: First, the fused signal is sliced ​​using a sliding window process, which can directly obtain or extract multiple signal segments of the fused signal. Then, a reconstruction algorithm is applied to reconstruct each signal segment to obtain the reconstructed signal of each signal segment, i.e., the reconstructed signal segment. Finally, based on the difference between each reconstructed signal segment and the corresponding signal segment, each reconstruction error element is obtained, which can be used to form the reconstruction error.

[0082] It is understood that the signal is time-series data, such as 100 sampling points per second. The fused signal can be sliced ​​using a sliding window in the time domain to obtain multiple signal segments. Here, the specific length of the sliding window and the specific number of signal segments can be optimized based on a comprehensive trade-off between the accuracy of actual signal fault detection and computational resources; this embodiment does not impose specific limitations on this. The specific calculation method for the reconstruction error element corresponding to each signal segment can employ methods such as Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE); this embodiment also does not impose specific limitations on this.

[0083] It should be noted that, given that the original signal usually has a large time scale, this solution improves the accuracy of signal processing and thus the accuracy of fault detection by dividing the time-domain signal into multiple time segments for independent processing.

[0084] In some embodiments, reconstructing based on each signal segment to obtain the reconstructed signal segment corresponding to each signal segment may include:

[0085] Each signal segment is input into the reconstruction model to obtain the reconstructed signal segments output by the reconstruction model. The number of convolutional modules included in the reconstruction model is determined based on the signal-to-noise ratio of the signal to be tested. The reconstruction model is trained based on the sample signal segments corresponding to the normal sample signal.

[0086] Specifically, the reconstruction model can be pre-trained as follows: First, collect normal sample signals from the target device and apply filtering to obtain corresponding filtered sample signals. Then, perform sliding window slicing on the fused sample signal obtained by fusing the normal and filtered sample signals to obtain individual sample signal segments. Finally, train the initial model based on each sample signal segment to obtain the language recognition model.

[0087] Based on this, the reconstruction model can be used to reconstruct each signal segment corresponding to the signal under test, thereby obtaining the reconstructed signal of each signal segment, i.e., the reconstructed signal segment.

[0088] It is understandable that, since the reconstruction model has learned the signal patterns of normal signals in advance, it can accurately reconstruct normal signals with relatively small reconstruction errors. However, if the signal segments input to the reconstruction model are fault signals, the output signal will deviate significantly from the original input signal. This deviation can then serve as data support for subsequent fault detection.

[0089] It should be noted that the reconstruction model training uses only normal signals, achieving high-sensitivity anomaly detection without relying on fault data or labeled data, demonstrating strong generalization ability. The reconstruction error calculated from the signal output by the trained reconstruction model characterizes the degree of variation of the abnormal signal relative to the normal signal. By introducing outlier detection algorithms from statistics to determine the outlier nature of the reconstruction error and providing a standard for anomaly identification, accurate fault detection can be achieved. Reconstructing the signal through signal slicing makes the model converge more easily, improving reconstruction accuracy. Furthermore, preprocessing the signal using filtering methods reduces interference from useless frequency bands, enhancing the learning effect of the reconstruction model and further improving fault detection accuracy.

[0090] Furthermore, to avoid increasing unnecessary training costs and computational overhead, and to improve the scalability of the algorithm, the number of convolutional modules included in the reconstruction model in this embodiment can be determined based on the signal-to-noise ratio (SNR) of the signal under test. It is understood that when the SNR of the signal under test is high, the signal is cleaner and more regular, making feature learning relatively easier; in this case, the reconstruction model can use a smaller number of convolutional modules. Conversely, when the SNR of the signal under test is low, feature learning is relatively more difficult; in this case, the reconstruction model can use a larger number of convolutional modules.

[0091] In some embodiments, in order to achieve high-fidelity signal reconstruction and improve detection robustness, the reconstruction model in this application embodiment may employ an autoencoder (AE).

[0092] Figure 2 This is a schematic diagram of the structure of the reconstruction model provided in the embodiments of this application, such as... Figure 2As shown, taking a reconstruction model with 3 convolutional modules as an example, this model can contain a structurally symmetrical encoder and decoder. The convolutional modules in the encoder and the deconvolutional modules in the decoder can be configured accordingly. The encoder maps the samples to a low-dimensional space to obtain the sample embedding vectors, and the decoder maps the embedding vectors to reconstructed samples, making the reconstructed samples as close as possible to the original samples. The state of a machine during normal operation is relatively stable, so normal signals often exhibit regular patterns, which the AE model can learn. AE can integrate network layers such as CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory) to enhance expressive power.

[0093] In some embodiments, step 140 may include:

[0094] If the proportion of outlier detection results in the outlier detection results corresponding to each reconstruction error element exceeds the target proportion, the fault detection result corresponding to the target device is determined to be a fault.

[0095] Specifically, the reconstruction error includes each reconstruction error element corresponding to each signal segment of the fused signal. Outlier detection is performed on each reconstruction error element to obtain the outlier detection results corresponding to each reconstruction error element. If the proportion of outliers in the outlier detection results exceeds the target proportion, the fault detection result corresponding to the target device can be determined as a fault. If the proportion of outliers in the outlier detection results does not exceed the target proportion, the fault detection result corresponding to the target device can be determined as no fault.

[0096] Here, the target ratio can be a pre-set fixed ratio, which can be set according to the actual application scenario, or it can be a ratio that is updated in real time based on the accuracy of fault detection. For example, if it is found that the current ratio misses a fault in the current fault detection scenario, the ratio can be lowered to improve the sensitivity of the algorithm for fault detection. This application embodiment does not specifically limit this. In practical applications, the ratio can be lowered in scenarios where timely maintenance is required or the downtime cost is small; the ratio can be increased in scenarios where equipment reliability is high or the downtime cost is high.

[0097] In some embodiments, step 130 may include:

[0098] Determine the median and the corresponding absolute median deviation of multiple reference error elements; the normal signal includes multiple normal signal segments, and the reference reconstruction error includes the reference error elements corresponding to the multiple normal signal segments;

[0099] Based on the difference between the reconstruction error and the median, and the absolute deviation of the median, the outlier value corresponding to the reconstruction error is determined;

[0100] Based on the comparison between the outlier values ​​corresponding to the reconstruction error and the target threshold, the outlier detection result corresponding to the reconstruction error is determined.

[0101] Specifically, for each reconstruction error element x in the reconstruction error, the Modified Z-score algorithm can be used to evaluate its value relative to the reference reconstruction error {x1, x2, ..., x}. n The outlier nature of} means that the reference reconstruction error here includes reference errors calculated from the reconstruction of multiple normal signal segments, i.e., multiple reference error elements. These multiple normal signal segments are obtained by sliding window slicing of the normal signal. The corrected Z-score calculation formula is as follows:

[0102]

[0103] in It is the median of the reference error elements. It is the median absolute deviation, which represents the median deviation of the reference error element from its median.

[0104] It should be noted that reconstruction errors usually do not conform to a normal distribution. However, in the embodiments of this application, the modified Z-score algorithm uses the median instead of the mean, which enhances the robustness of the algorithm and also performs well when the data does not conform to a normal distribution.

[0105] The corrected Z-score calculated for each reconstruction error element is compared with the set threshold. If Z > threshold, the outlier detection result corresponding to the reconstruction error element is determined to be an outlier; if Z ≤ threshold, the outlier detection result corresponding to the reconstruction error element is determined to be a non-outlier.

[0106] The specific number of normal signal segments used here can be optimized based on the accuracy of actual signal fault detection and a comprehensive trade-off of computing resources. This application embodiment does not impose a specific limitation on this.

[0107] The target threshold can be a pre-set fixed threshold, which can be set according to the actual application scenario, or it can be an updatable threshold. For example, if it is found that the current threshold has a missed fault detection in the current fault detection scenario, the threshold can be lowered to improve the sensitivity of the algorithm for fault detection. This application embodiment does not make specific limitations on this.

[0108] In some embodiments, this application proposes a fault detection method. This method only requires the vibration signal of the target device when it is operating normally as a reference. It learns the pattern of the normal signal through an autoencoder and uses an outlier detection algorithm such as Z-score correction to determine whether the reconstruction error of the signal under test after the autoencoder is an outlier relative to the reconstruction error of the reference signal. Thus, it determines whether the signal under test contains a fault.

[0109] Most related technologies rely on fault data to train classification models, while those that do not use fault data use single-classification methods to model normal data. This application's embodiment uses the similarity between the tested signal and a normal signal as the criterion for fault identification, thus avoiding reliance on fault data. The reconstruction error of the alternating image (AE) is used to characterize the degree of variation of the abnormal signal relative to the normal signal. An outlier detection algorithm from statistics is introduced to determine the outlier nature of the reconstruction error, providing a criterion for anomaly identification. Both the AE and the outlier detection algorithm used in this architecture are modular methods. The network layers in the AE structure can be customized, and the outlier detection algorithm can employ modified Z-scores, interquartile range (ICM)-based methods, etc.

[0110] To enhance the learning performance of the autoencoder, a filtering method is used to preprocess the signal and reduce interference from useless frequency bands. This method does not require the collection or labeling of fault samples, making it suitable for data-scarce situations in real-world industrial scenarios. It features a modular structure, supports customized model building and flexible threshold settings, adapts to different equipment and operating conditions, and can handle multi-channel, high-dimensional, and multi-source signals, exhibiting good versatility and robustness.

[0111] This method includes several steps: data preprocessing, model training, and outlier detection.

[0112] (1) Data preprocessing: Assuming the rotational speed frequency of the equipment is f, a narrowband bandpass filter is used to extract the components of the original signal near frequency f, and reconstruct a new signal. The reconstructed signal and the original signal are spliced ​​together as a new channel.

[0113] (2) Model Training: An autoencoder model is used for training, employing only normal signals. The trained AE model can accurately reconstruct normal signals. A complete signal is divided into multiple segments using a sliding window, with each segment serving as a sample input to the model. AE can integrate CNN, LSTM, and other network layers to enhance its expressive power.

[0114] (3) Outlier Detection: After model training, a portion of normal signal samples are randomly selected as reference samples. These reference samples are input into the model to calculate the reconstruction error, which is called the reference reconstruction error. For the signal under test, the same denoising method and sliding window slicing are used to obtain multiple signal segments as samples. A portion of these segments is extracted and input into the model to calculate the reconstruction error, which is also called the test reconstruction error. The modified Z-score is used to evaluate the outlier nature of the test reconstruction error relative to the reference reconstruction error. If 80% or more (this percentage is configurable) of the signal segments under test are judged as outliers, the signal under test is marked as having an anomaly, i.e., the fault detection result of the target device is a fault.

[0115] The embodiments of this application solve the problems of strong dependence on fault data, insufficient model generalization ability and lack of flexibility of related methods through the above design. The modular design makes each component independent and replaceable and easy to expand, which can be widely adapted to different industrial objects and signal types.

[0116] It should be noted that, compared to fault detection methods using boundary discriminant classifiers in related technologies, which lack adjustable quantitative indicators and cannot set different thresholds and warning levels for different scenarios, the parameters mentioned in the embodiments of this application can all be changed. For example, the number of normal signal segments n used to determine the reference reconstruction error, the number of signal segments m of the test signal used to determine the reconstruction error, the threshold for judging outliers, and the proportion of outliers used to judge faults. n and m can be set according to a comprehensive trade-off between fault detection accuracy and computational resources, or according to the signal-to-noise ratio of the signal. For example, n and m can be appropriately reduced when the signal-to-noise ratio is high. If it is found that the default threshold has missed alarms in the current scenario (not sensitive enough to faults), the detection threshold can be lowered to make the algorithm more likely to issue an alarm.

[0117] In some embodiments, this application specifically relates to intelligent condition monitoring and fault diagnosis technology for industrial rotating equipment, proposing a fault detection method for rotating machinery vibration signals based on a self-encoder, which combines signal filtering, reconstruction error analysis, sliding window strategy, outlier detection and other means to achieve fault detection and early warning for industrial equipment.

[0118] Taking the publicly available wind turbine gearbox dataset from the New Energy Laboratory as an example, this paper illustrates the implementation process of the fault detection method:

[0119] (1) Data preprocessing

[0120] The fault detection method requires time-series data collected from one or more vibration sensors. Different measurement points constitute multiple signal channels. For example, there is one sensor at each end of the engine and at the base. Each sensor samples 100 times per second, resulting in a 100x3 matrix of data per second. This data contains three channels, with each "channel" being one column. For each channel, a narrowband bandpass filter is used to filter the signal, retaining only the components within a frequency band centered at frequency f and with a radius of width (this value is configurable), where f is the rotational speed frequency of the target equipment. The filtered signal is then concatenated with the original signal, doubling the number of channels. Taking width=1 as an example, the time-domain waveform comparison before and after filtering provided in this embodiment is shown below. Figure 3 As shown.

[0121] Related techniques typically involve extracting features or spectrograms from the original signal, or directly inputting the original signal into a model (end-to-end method). This application's embodiment concatenates the original signal and the filtered signal, effectively utilizing features from both sources. The filtered signal is used because filtering makes the fault-related aspects of the signal more prominent.

[0122] (2) Model Training

[0123] A sliding window is used to extract samples from the spliced ​​signal, and each sample still contains multiple channels.

[0124] An autoencoder model is trained using truncated normal signal samples. This model comprises a structurally symmetric encoder and decoder. The encoder maps the samples to a low-dimensional space, obtaining the sample's embedding vector, while the decoder maps the embedding vector to reconstructed samples, ensuring the reconstructed samples are as close as possible to the original samples. Since the state of a machine during normal operation is relatively stable, normal signals often exhibit regular patterns, which the autoencoder model can learn.

[0125] The structure of AE in the embodiments of this application is as follows: Figure 2 As shown, in practical applications, network layers can be added or reduced. For example, for signals with simpler patterns, the encoder and decoder only need two convolutional modules.

[0126] (3) Outlier detection

[0127] After the model training is complete, n=10 normal signal samples (this value is configurable) are randomly selected as reference samples. The reference samples are input into the model to calculate the reconstruction error, which is called the reference sample error, or the reference reconstruction error in step 130. Since the reference samples conform to the rules learned by the model, the reference sample error should be relatively small.

[0128] For the signal under test, the same denoising method and sliding window slicing are used to obtain multiple signal segments as samples. From these, m = 5 (this value is configurable) segments are extracted and input into the model to calculate the reconstruction error, i.e., the error of the test sample. For each test sample error x, a corrected Z-score is used to evaluate its error relative to the reference sample error {x1, x2, ..., x}. n The outlier property of}. The modified Z-score calculation formula is as follows:

[0129]

[0130] in It is the median of the reference sample error. It is the median absolute deviation, which represents the median deviation of a data point from its median.

[0131] The corrected Z-score calculated from the error of the sample to be tested is compared with the set threshold. If Z > threshold, the error of the sample to be tested is determined to be abnormal, i.e., an outlier. This threshold is usually universal, for example, it can be set to the threshold of 3.5 in the field of statistics.

[0132] It should be noted that AE itself is not a single-classification model. In this embodiment, the reconstruction error of AE is used to reflect the fault, that is: the error of the reference sample reconstructed by the autoencoder model is used to evaluate the outlierness of the error of the test sample reconstructed by the autoencoder model, and then the outlier value is compared with the set threshold to determine whether the test sample is abnormal, so that AE can perform the same function as a single-classification model.

[0133] If 80% or more (this percentage is configurable) of the segments in the signal under test are identified as outliers, the signal is marked as abnormal. The lower this percentage, the more sensitive the fault detection method is to anomalies. In practical applications, this percentage can be lowered for scenarios requiring timely maintenance or with low downtime costs; and higher for scenarios requiring high equipment reliability or with high downtime costs.

[0134] The fault detection method provided in this application can be executed by a fault detection device. This application uses the example of a fault detection device executing the fault detection method to illustrate the fault detection device provided in this application.

[0135] This application also provides a fault detection device.

[0136] Figure 4 This is a schematic diagram of the fault detection device provided in the embodiments of this application, as shown below. Figure 4 As shown, the fault detection device includes: a signal acquisition module 410, an error determination module 420, an outlier detection module 430, and a fault detection module 440.

[0137] Signal acquisition module 410 is used to acquire the signal to be tested from the target device;

[0138] Error determination module 420 is used to determine the reconstruction error corresponding to the signal under test based on the signal obtained by reconstructing the signal under test;

[0139] The outlier detection module 430 is used to detect outliers in the reconstruction error based on a reference reconstruction error, and obtain an outlier detection result corresponding to the reconstruction error; the reference reconstruction error is determined based on the signal reconstructed from the normal signal corresponding to the signal under test.

[0140] The fault detection module 440 is used to determine the fault detection result corresponding to the target device based on the outlier detection result corresponding to the reconstruction error.

[0141] According to the fault detection device provided in the embodiments of this application, the signal under test is reconstructed, and the reconstruction error is calculated. Then, an outlier detection algorithm in statistics is introduced to determine the outlierness of the reconstruction error relative to the reference reconstruction error calculated from the normal signal, and a standard for judging anomalies is given. This avoids dependence on fault data and improves the accuracy, generalization and robustness of fault detection. In addition, the reconstruction algorithm and the outlier detection algorithm can be flexibly configured and easily expanded, and can be widely adapted to fault detection and early warning of different industrial objects and signal types.

[0142] The fault detection device in this application embodiment can be an electronic device or a component within an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal or other devices besides a terminal. For example, the electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc. It can also be a server, network attached storage (NAS), personal computer (PC), television (TV), ATM, or self-service machine, etc. This application embodiment does not specifically limit the specific type of device.

[0143] The fault detection device in this application embodiment can be a device with an operating system. This operating system can be a Microsoft (Windows) operating system, an Android operating system, an iOS operating system, or other possible operating systems; this application embodiment does not specifically limit the specific operating system.

[0144] The fault detection device provided in this application embodiment can achieve... Figures 1 to 3 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0145] In some embodiments, such as Figure 5 As shown, this application embodiment also provides an electronic device 500, including a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the program is executed by the processor 501, it implements the various processes of the above-described fault detection method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0146] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0147] This application also provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described fault detection method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0148] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0149] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described fault detection method.

[0150] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0151] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described fault detection method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0152] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0153] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0154] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the related technology, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0155] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0156] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0157] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A fault detection method, characterized in that, include: Acquire the signal to be tested from the target device; Based on the signal reconstructed from the signal under test, determine the reconstruction error corresponding to the signal under test; Based on the reference reconstruction error, outlier detection is performed on the reconstruction error to obtain the outlier detection result corresponding to the reconstruction error. The reference reconstruction error is determined based on the signal obtained by reconstructing the normal signal corresponding to the signal under test. Based on the outlier detection results corresponding to the reconstruction error, the fault detection results corresponding to the target device are determined.

2. The fault detection method according to claim 1, characterized in that, The process of determining the reconstruction error corresponding to the signal under test based on the reconstructed signal includes: Based on the operating frequency of the target device, the signal to be tested is bandpass filtered to obtain a filtered signal. The filtered signal is fused with the signal to be tested to obtain a fused signal; The reconstruction error is determined based on the signal obtained by reconstructing the fused signal.

3. The fault detection method according to claim 2, characterized in that, The determination of the reconstruction error based on the signal reconstructed from the fused signal includes: The fused signal is subjected to sliding window processing to obtain multiple signal segments of the fused signal; Reconstruction is performed based on each of the signal segments to obtain the reconstructed signal segments corresponding to each of the signal segments; Based on each of the reconstructed signal segments and the corresponding signal segments, the reconstructed error includes each reconstructed error element.

4. The fault detection method according to claim 3, characterized in that, The reconstruction based on each of the signal segments to obtain the reconstructed signal segments corresponding to each of the signal segments includes: Each of the signal segments is input into the reconstruction model to obtain each of the reconstructed signal segments output by the reconstruction model; the number of convolutional modules included in the reconstruction model is determined based on the signal-to-noise ratio of the signal to be tested, and the reconstruction model is trained based on the sample signal segments corresponding to the sample normal signal.

5. The fault detection method according to claim 3, characterized in that, The step of determining the fault detection result corresponding to the target device based on the outlier detection result corresponding to the reconstruction error includes: If the proportion of outlier detection results in the outlier detection results corresponding to each of the reconstruction error elements exceeds the target proportion, the fault detection result corresponding to the target device is determined to be a fault.

6. The fault detection method according to any one of claims 1 to 5, characterized in that, The step of performing outlier detection on the reconstruction error based on the reference reconstruction error to obtain the outlier detection result corresponding to the reconstruction error includes: The median and the corresponding absolute deviation of the median are determined from multiple reference error elements; the normal signal includes multiple normal signal segments, and the reference reconstruction error includes the reference error elements corresponding to the multiple normal signal segments; Based on the difference between the reconstruction error and the median, and the absolute deviation of the median, the outlier value corresponding to the reconstruction error is determined; Based on the comparison between the outlier value corresponding to the reconstruction error and the target threshold, the outlier detection result corresponding to the reconstruction error is determined.

7. A fault detection device, characterized in that, include: The signal acquisition module is used to acquire the signal to be tested from the target device. An error determination module is used to determine the reconstruction error corresponding to the signal under test based on the signal reconstructed from the signal under test. An outlier detection module is used to detect outliers in the reconstruction error based on a reference reconstruction error, and to obtain the outlier detection result corresponding to the reconstruction error. The reference reconstruction error is determined based on the signal obtained by reconstructing the normal signal corresponding to the signal under test. The fault detection module is used to determine the fault detection result corresponding to the target device based on the outlier detection result corresponding to the reconstruction error.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the fault detection method as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fault detection method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the fault detection method as described in any one of claims 1-6.

Images