A fault detection method, apparatus and device

Abstract

The application provides a fault detection method, device and equipment. The fault detection method provided by the application comprises: acquiring acoustic data in the running process of a to-be-tested device; performing feature extraction on the acoustic data to obtain a global feature matrix corresponding to the acoustic data; wherein the global feature matrix comprises multidimensional features; constructing a relationship graph corresponding to the global feature matrix; wherein each feature in the global feature matrix is taken as a node, a directed connection relationship between any two features is taken as an edge, and a correlation coefficient between the two features is taken as a weight of the edge; and a pre-trained recognition model is used to perform fault recognition on the relationship graph to obtain a fault recognition result of the to-be-tested device. The fault detection method, device and equipment provided by the application can fully utilize the dependency relationship and relationship strength between features for recognition, effectively improving the accuracy and reliability of fault detection.

Description

Technical Field

[0001] This application relates to the field of fault detection technology, and in particular to a fault detection method, apparatus and equipment. Background Technology

[0002] In recent years, with the rapid development of sensing technology and big data analytics, more efficient and accurate fault detection of industrial equipment has become possible. By using multi-source sensors to collect real-time status information of the equipment under test during operation, and combining it with advanced analysis techniques, potential faults can be identified in a timely manner when abnormal signs appear, thereby effectively ensuring the continuity, safety, and efficiency of industrial production.

[0003] Sound is an inherent property of equipment operation. When equipment experiences wear, loosening, or other abnormalities, its acoustic patterns often undergo identifiable changes. Therefore, by collecting and analyzing these sound signals, continuous monitoring of equipment operating status and early fault identification can be achieved, providing a non-contact detection method for industrial systems. However, in complex industrial environments, sound signals are easily affected by noise. How to improve the accuracy of identification under conditions of multiple interferences has become a current research hotspot. Summary of the Invention

[0004] In view of this, this application provides a fault detection method, apparatus, and device to improve the accuracy of fault detection.

[0005] The first aspect of this application provides a fault detection method, the fault detection method comprising:

[0006] Acquire acoustic data during the operation of the device under test;

[0007] Feature extraction is performed on the acoustic data to obtain a global feature matrix corresponding to the acoustic data; wherein, the global feature matrix includes multi-dimensional features;

[0008] Construct a relational graph corresponding to the global feature matrix; wherein, the relational graph uses each feature dimension in the global feature matrix as a node, the directed connection relationship between any two features as an edge, and the correlation coefficient between the two features as the weight of the edge;

[0009] The fault identification results of the device under test are obtained by using a pre-trained recognition model to perform fault identification on the relationship graph.

[0010] A second aspect of this application provides a fault detection device, which includes an acquisition module, an extraction module, a construction module, and a recognition model; wherein...

[0011] The acquisition module is used to acquire acoustic data during the operation of the device under test;

[0012] The extraction module is used to extract features from the acoustic data to obtain a global feature matrix corresponding to the acoustic data; wherein, the global feature matrix includes multi-dimensional features;

[0013] The construction module is used to construct a relational graph corresponding to the global feature matrix; wherein, the relational graph uses each feature dimension in the global feature matrix as a node, the directed connection relationship between any two features as an edge, and the correlation coefficient between the two features as the weight of the edge;

[0014] The identification module is used to perform fault identification on the relationship graph using a pre-trained identification model to obtain the fault identification result of the device under test.

[0015] A third aspect of this application provides a fault detection 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 steps of any of the fault detection methods provided in the first aspect of this application.

[0016] The fault detection method, apparatus, and equipment provided in this application acquire acoustic data during the operation of the device under test, extract features from the acoustic data to obtain a global feature matrix corresponding to the acoustic data, and further construct a relational graph corresponding to the global feature matrix. A pre-trained recognition model is then used to identify faults in this relational graph, resulting in a fault identification result for the device under test. By constructing a relational graph from the global feature matrix, the dependencies and correlation strengths between features in each dimension can be explicitly characterized. This allows the recognition model to consider not only single-dimensional information when processing features but also to fully utilize the dependencies and correlation strengths between features for identification, effectively improving the accuracy and reliability of fault detection.

[0017] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0019] Figure 1 A flowchart of Embodiment 1 of the fault detection method provided in this application;

[0020] Figure 2 This is a partial schematic diagram illustrating a constructed relationship graph as shown in an exemplary embodiment of this application;

[0021] Figure 3 A flowchart of Embodiment 2 of the fault detection method provided in this application;

[0022] Figure 4 This is a schematic diagram illustrating an exemplary embodiment of the recognition model of this application;

[0023] Figure 5 A flowchart of Embodiment 3 of the fault detection method provided in this application;

[0024] Figure 6 This is a schematic diagram of a circular storage queue shown in an exemplary embodiment of this application;

[0025] Figure 7 A hardware structure diagram of a fault detection device, which is a fault detection apparatus, shown as an exemplary embodiment of this application;

[0026] Figure 8 This is a schematic diagram of the structure of a first embodiment of the fault detection device provided in this application. Detailed Implementation

[0027] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0028] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0029] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0030] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0031] Figure 1 This is a flowchart of an embodiment of the fault detection method provided in this application. Please refer to... Figure 1The method provided in this embodiment may include:

[0032] S101. Acquire acoustic data during the operation of the device under test.

[0033] It should be noted that the fault detection method and fault detection device provided in this embodiment are applied to fault detection equipment, which may be a server or the like. In this embodiment, it is not limited to such equipment.

[0034] The device under test (DUT) can be any type of industrial equipment; this embodiment does not limit its application. Specifically, an acquisition device can be used to collect acoustic data during the operation of the DUT, and then the acoustic data can be obtained from the acquisition device. Alternatively, the acquisition device can collect the acoustic data during the operation of the DUT in real time and send it to a fault detection device in real time. The fault detection device stores the received acoustic data in a storage area, and then the acoustic data during the operation of the DUT can be obtained from the storage area. For example, in one embodiment, the acquired acoustic data may include at least one sound frame.

[0035] Specifically, for example, in one embodiment, a microphone is used to collect acoustic signals during the operation of the device under test in real time. The microphone can save a sound signal of a specified duration or a specified amount of data as a sound frame. Furthermore, when fault detection is required, the fault detection device acquires at least one sound frame from the microphone.

[0036] The specified duration or data volume is set according to actual needs. In this embodiment, for example, the specified duration can be 5 minutes, 10 minutes, 20 minutes, etc. Similarly, the specified data volume can be 5KB, etc.

[0037] It should be noted that two adjacent audio frames may or may not overlap; this embodiment does not impose any restrictions on this.

[0038] S102. Perform feature extraction on the acoustic data to obtain a global feature matrix corresponding to the acoustic data; wherein, the global feature matrix includes multi-dimensional features.

[0039] In practice, as described above, when the acoustic data includes a single sound frame, it can be segmented into multiple sound frames. Feature extraction is then performed on each sound frame to obtain its local feature vector. Finally, the local feature vectors of all sound frames are concatenated in chronological order to obtain the global feature matrix.

[0040] When acoustic data includes multiple sound frames, features can be directly extracted from each sound frame to obtain its local feature vector. Then, the local feature vectors of all sound frames can be concatenated in chronological order to obtain a global feature matrix. Alternatively, when acoustic data includes multiple sound frames, each sound frame can be further segmented into multiple sound frames, and features can then be extracted from each sound frame.

[0041] Optionally, in one possible implementation, the acoustic data includes a preset number of sound frames. Specifically, the preset number of sound frames can be obtained by the acquisition device automatically performing frame segmentation processing on the sound signal after acquisition, or by the fault detection device performing frame segmentation processing on the acquired sound signal after the acquisition device sends it to the fault detection device. In this embodiment, this is not limited. Furthermore, referring to the preceding description, frame segmentation processing can include non-overlapping frame segmentation and overlapping frame segmentation. In this embodiment, this is not limited.

[0042] Furthermore, the specific value of the preset number of frames is set according to actual needs, and is not limited in this embodiment. For ease of explanation, the preset number of frames is denoted as N frames.

[0043] In practical implementation, the specific number of multidimensional features and the specific form of each feature are set according to actual needs when extracting features; this embodiment does not limit them. Furthermore, in one possible implementation, the multidimensional features simultaneously include time-domain features and frequency-domain features. For example, in one possible implementation, the multidimensional features include 12 dimensions, specifically 8 time-domain features and 4 frequency-domain features. Further, the 8 time-domain features are: short-time energy, short-time zero-crossing rate, root mean square, kurtosis, skewness, peak factor, impulse factor, and margin factor; the 4 frequency-domain features are: spectral centroid, spectral bandwidth, spectral kurtosis, and band energy.

[0044] It should be noted that short-time energy represents the sum of the squares of the amplitudes of all sound signals in each sound frame. This time-domain feature characterizes the total energy level of the sound signal and can be used to distinguish between audible and silent segments, and monitor energy fluctuations during the operation of the device under test. Short-time zero-crossing rate is used to count the number of times the amplitude of all sound signals in each sound frame crosses the horizontal axis. Root mean square (RMS) is a measure of average power, reflecting the average energy intensity of the sound frame. Kurtosis, as the fourth-order central moment of the sound signal, characterizes the heaviness of the tail of the signal probability density distribution and is extremely sensitive to impulsive components in the signal. Skewness, as the third-order central moment of the sound signal, reflects the degree of asymmetry between the positive and negative amplitudes of the waveform. Crest factor is the ratio of the peak value of the sound signal to the RMS, used to assess whether there are prominent impulsive peaks in the signal. Impulse factor is the ratio between the maximum absolute value of the sound signal and the mean absolute value. Margin factor is the ratio between the maximum value of the signal and the average of the square roots of the signal's absolute value, quantifying the significance of possible abnormal impulsive peaks in the signal relative to the overall normal vibration level.

[0045] Furthermore, the spectral centroid characterizes the central location of sound frequency energy, reflecting the overall shift trend of the spectrum. Spectral bandwidth measures the degree of dispersion of the spectrum around the centroid, used to distinguish spectral shape. Spectral kurtosis describes the sharpness of the spectral energy distribution and is a fourth-order statistic in the frequency domain. Bandwidth energy characterizes the total energy of a signal within a specific band of interest, used to quantify the total energy of a particular frequency band.

[0046] It should be noted that the specific implementation methods and principles for extracting each time-domain feature and each frequency-domain feature can be found in the descriptions in relevant technologies, and will not be repeated here. For ease of explanation, these 12 features are denoted as Q1, Q2, ..., Q12 in sequence.

[0047] Furthermore, after extracting the local feature vectors of each audio frame, the local feature vectors of all audio frames are concatenated in chronological order to obtain the global feature matrix. Thus, each feature dimension contains a predetermined number of feature values ​​(equal to a predetermined number of frames), meaning that each feature value corresponds to a different time point. In other words, each feature dimension includes feature values ​​from multiple different time points.

[0048] In practice, each local feature vector can be treated as a row vector, and each local feature vector can be arranged row by row to obtain the global feature matrix. Alternatively, each local feature vector can be treated as a column vector, and each local feature vector can be arranged column by column to obtain the global feature matrix.

[0049] The following example illustrates how to "arrange each local feature vector as a row vector, and then arrange each local feature vector row by row to obtain the global feature matrix". In this case, combining the example above, when the multi-dimensional features include 12 dimensions, for each audio frame, its local feature vector is a 1... The row vectors are further processed by arranging the local feature vectors of each audio frame as a row in chronological order (frame order) to obtain the global feature matrix. This global feature matrix is ​​an N-dimensional vector matrix. The global feature matrix is ​​a feature matrix that accumulates over time.

[0050] S103. Construct a relational graph corresponding to the global feature matrix; wherein, the relational graph uses each feature dimension in the global feature matrix as a node, the directed connection relationship between any two features as an edge, and the correlation coefficient between the two features as the weight of the edge.

[0051] It should be noted that a relation graph includes node combinations, edge sets, and weight matrices. For example, Figure 2 This is a partial schematic diagram illustrating the constructed relationship graph as shown in an exemplary embodiment of this application. Please refer to... Figure 2 The node set consists of multiple nodes, each representing a feature in one dimension of the global feature matrix. Furthermore, the edge set consists of multiple edges, each representing a directed connection between two features. The weight matrix records the correlation coefficient between any two features.

[0052] Optionally, in one possible implementation, based on the above embodiments, constructing the relational graph corresponding to the global feature matrix includes:

[0053] (1) Extract each dimension feature from the global feature matrix, and use each dimension feature as a node to determine the set of all nodes as the node set.

[0054] Referring back to the example above, when the global feature matrix contains 12 features, it comprises 12 nodes. Similarly, when each row vector represents a local feature vector, each column vector corresponds to a one-dimensional feature, meaning each node corresponds to a column vector of the global feature matrix. The node set can be represented as: ,in, Indicates the first 3D features The corresponding node is an N The vector, Take numbers 1 to 12.

[0055] (2) For any two-dimensional feature in the global feature matrix, construct the directed edges corresponding to the two-dimensional feature, and determine the set of all directed edges as the edge set.

[0056] Specifically, in combination Figure 2 For any two features in a multi-dimensional feature matrix, a directed connection is established between them as an edge. For example, for the first and second features, there exist directed edges from the first feature to the second feature, and from the second feature to the second feature. Combining this example, when the global feature matrix contains 12 nodes, the edge set includes 12 nodes. There are 11 edges. For ease of explanation, let the set of edges be denoted as B, where the element Bij represents the edge of the first edge. A directed edge pointing from the j-th dimension feature to the j-th dimension feature, i.e., a node. Pointing to node The directed edge.

[0057] (3) For any two-dimensional features in the global feature matrix, determine the correlation coefficient between the two-dimensional features.

[0058] In practice, for any two-dimensional feature, the Pearson correlation coefficient, Spearman rank correlation coefficient, etc., can be calculated between them. In this embodiment, no limitation is imposed.

[0059] (4) The correlation coefficient between any two-dimensional features is determined as the weight of the directed edge corresponding to the two-dimensional features, and the matrix formed by the weights of all directed edges is determined as the weight matrix.

[0060] Specifically, after calculating the correlation coefficient between any two-dimensional features, it is used as the weight of the directed edge corresponding to that two-dimensional feature, resulting in a weight matrix. For example, for the The correlation coefficient between the j-th and j-th dimension features can be calculated. And thus take it as the first The weight of the directed edge pointing to the j-th feature.

[0061] Furthermore, for ease of explanation, in this embodiment, the weight matrix is ​​denoted as W.

[0062] (5) The combination of the node set, the edge set, and the weight matrix is ​​determined as the relation graph.

[0063] Based on the preceding description, it can be understood that the relational graph can be represented as: G = .

[0064] It should be noted that in this embodiment, by abstracting each feature dimension into nodes and constructing weighted directed edges using correlation coefficients, the potential dependencies, influence directions, and correlation strength between features can be quantified. In subsequent processing, the recognition model can distinguish between strong and weak correlations between features based on the edge weights, thereby highlighting the interaction relationships between key features. Thus, with the help of the relationship graph, the originally scattered high-dimensional feature data is organized into structured topological information, enhancing the recognition model's ability to identify feature interaction patterns and improving the overall accuracy and robustness of fault detection.

[0065] S104. Use the pre-trained recognition model to perform fault identification on the relationship graph to obtain the fault identification result of the device under test.

[0066] It should be noted that the specific structure of the recognition model is set according to actual needs, and is not limited in this embodiment. For example, in one possible implementation, it can be a graph neural network. Furthermore, the recognition model is a pre-trained model, and its specific training process can be found in the descriptions in related technologies, which will not be repeated here.

[0067] In this step, the constructed relationship graph is input into a pre-trained recognition model. This model extracts and classifies features from the graph based on the characteristic relationships represented by nodes, edges, and their weights, thereby determining the current fault type of the device under test. Finally, the recognition model outputs the corresponding fault identification result, achieving automatic identification of the operating status of the device under test.

[0068] The method provided in this embodiment acquires acoustic data during the operation of the device under test, extracts features from the acoustic data to obtain a global feature matrix corresponding to the acoustic data, and further constructs a relational graph corresponding to the global feature matrix. Then, a pre-trained recognition model is used to identify faults in the relational graph to obtain the fault identification result of the device under test. In this way, by constructing the global feature matrix into a relational graph, the dependency relationship and correlation strength between each dimension feature can be explicitly characterized. This allows the recognition model to not only consider single-dimensional information when processing features, but also make full use of the dependency relationship and correlation strength between features for identification, which can effectively improve the accuracy and reliability of fault detection.

[0069] Figure 3 This is a flowchart of Embodiment 2 of the fault detection method provided in this application. Please refer to... Figure 3 In one possible implementation, based on the above embodiments, determining the correlation coefficient between any two-dimensional features in the global feature matrix may include:

[0070] S301. For each dimension of the global feature matrix, normalize each element in that dimension to obtain the standardized feature matrix corresponding to the global feature matrix.

[0071] In practice, the mean and standard deviation of each feature dimension can be calculated first. Then, each element in the feature dimension can be normalized according to the mean and standard deviation of each feature dimension to obtain a standardized feature matrix.

[0072] Combining the previous example, when the global feature matrix is ​​obtained by arranging each local feature vector in a row-by-row manner, each feature dimension corresponds to a column vector in the global feature matrix. In this case, the global feature matrix is ​​processed column by column. For each column, the mean and standard deviation of all elements contained in the column are calculated, and then the mean and standard deviation are used to normalize each element in the column.

[0073] S302. Calculate the covariance matrix of the standardized feature matrix, and determine the Pearson correlation coefficient between any two features based on the covariance matrix.

[0074] In practice, the covariance matrix can be calculated using the following formula:

[0075] ;

[0076] Among them, the elements in the covariance matrix Indicates the first The covariance of the j-th feature and the j-th feature.

[0077] It is understandable that, in the example above, the global feature matrix has a dimension of N. At 12 o'clock, the dimension of the covariance matrix is ​​12. 12.

[0078] Furthermore, the Pearson correlation coefficient can be calculated using the following formula:

[0079] ;

[0080] in, Indicates the first The Pearson correlation coefficient between the j-th dimension feature and the j-th dimension feature.

[0081] S303. The absolute value of the Pearson correlation coefficient between any two-dimensional features is determined as the correlation coefficient between the two-dimensional features.

[0082] In this step, for any two-dimensional features, the absolute value of the Pearson correlation coefficient between these two features is determined as its correlation coefficient.

[0083] The method provided in this embodiment first normalizes the features of each dimension in the global feature matrix to obtain a standardized feature matrix, and then calculates the covariance matrix based on the standardized feature matrix, thereby obtaining the Pearson correlation coefficient between any two-dimensional features. This enables a unified scale measurement of the correlation between features of different dimensions and numerical ranges. In this way, it not only avoids the bias caused by the difference in the amplitude of the original features on the correlation calculation results, but also accurately characterizes the degree of mutual correlation between two-dimensional features, providing a reliable foundation for the subsequent construction of feature maps.

[0084] Optionally, in one possible implementation, Figure 4 This is a schematic diagram illustrating an exemplary embodiment of the recognition model of this application. Please refer to... Figure 4 The recognition model includes a spatiotemporal attention module and a classification module; the spatiotemporal attention module includes at least one spatiotemporal attention layer; wherein,

[0085] Each spatiotemporal attention layer is used to perform spatial aggregation and temporal compression on the feature matrix currently represented by the node set based on the edge set and the weight matrix, so as to obtain the spatiotemporal correlation feature matrix corresponding to the global feature matrix.

[0086] The classification module is used to perform fault identification on the spatiotemporal correlation feature matrix to obtain the fault identification result.

[0087] It should be noted that the number of spatiotemporal attention layers included in the spatiotemporal attention module is set according to actual needs, and is not limited in this embodiment.

[0088] For further details, please refer to [link / reference]. Figure 4 Each spatiotemporal attention layer can include a spatial attention layer and a temporal convolutional layer. Furthermore, the spatial attention layer is used to achieve spatial aggregation, and the temporal convolutional layer is used to achieve temporal compression.

[0089] It should be noted that spatial aggregation is used to weight and converge the features of each node based on the topological relationships described by the edge set, in order to extract the spatial correlation between nodes. Temporal compression, on the other hand, is used to reduce the dimensionality of the features of nodes that change over time, in order to extract dynamic associations across time. Through the joint processing of spatial aggregation and temporal compression, the spatiotemporal attention layer can extract spatiotemporal correlation features from the original features that simultaneously reflect spatial dependencies and temporal evolution.

[0090] Optionally, in one possible implementation, performing spatial aggregation and temporal compression on the feature matrix currently represented by the node set based on the edge set and the weight matrix includes:

[0091] (1) For each node in the node set, determine the set of neighboring nodes corresponding to that node based on the edge set.

[0092] Understandably, see Figure 4 For the first spatiotemporal attention layer, the feature matrix represented by the current node set is the initial global feature matrix. For subsequent spatiotemporal attention layers, the feature matrix represented by the current node set is the spatiotemporal correlation feature matrix output by the previous spatiotemporal attention layer.

[0093] Specifically, for a given node, if an edge exists between it and another node, then that other node is a neighbor node of that node. (See reference...) Figure 2 For nodes Its set of neighboring nodes includes: , , , , ... The neighbor node sets of other nodes are similar, and will not be described again in this embodiment.

[0094] (2) For each edge in the edge set, determine the attention coefficient of the edge according to the weight of the edge and the weight of the associated edge of the edge; wherein the associated edge is the edge between the starting node of the edge and each of the neighboring nodes of the starting node.

[0095] Specifically, for example, for any edge in the edge set The starting node of this edge is The end node is The associated edge of this edge is and The edges between each of the neighboring nodes. For example, for an edge The associated edges of this edge include , B15, etc.

[0096] Furthermore, in this step, the weights of each edge can be obtained from the weight matrix, and then the attention coefficient of each edge can be calculated according to the following formula:

[0097] ;

[0098] in, Representing an edge Attention coefficient; Indicates the activation function; Represents a node The set of neighboring nodes, which contains nodes All neighboring nodes, Representing an edge Weights of edges Refers to nodes The edges between this node and its neighboring nodes, i.e., edges The associated edges.

[0099] (3) For each node in the node set, the current features of the neighboring nodes of the node are weighted according to the attention coefficient of the edge between the node and the neighboring nodes of the node to obtain the updated feature increment of the node.

[0100] Specifically, nodes can be determined using the following formula. Update feature increment:

[0101] ;

[0102] in, Represents a node The update feature increment; Represents a pooling function; Represents a node and the neighboring nodes of this node Attention coefficients of the edges between them Represents a node neighboring nodes The current characteristics.

[0103] Combining the examples above, for instance, for the node... The update feature increment of the node can be determined using the following formula:

[0104] ;

[0105] (4) Determine the update space features of the node based on its current features and the update feature increment of the node.

[0106] Specifically, the updated spatial features can be determined using the following formula:

[0107] ;

[0108] in, For nodes The update space features; A is the parameter matrix of the recognition model, which is obtained through training; Indicates the join function; Represents a node Current characteristics; For nodes The update feature increment.

[0109] Referring to the preceding description, it can be understood that steps (1) to (4) constitute the spatial aggregation process. Spatial aggregation does not change the dimension of the feature corresponding to a node; for a node, its original feature has the same dimension as the dimension of its updated spatial feature. Combining the previous example, for a node, its corresponding feature is N. A vector of 1, after spatial aggregation, still updates its spatial features to N. A vector of 1.

[0110] As described above, spatial aggregation updates the features of a node by utilizing the correlation between the node and other nodes. This introduces neighborhood information within a local scope, enabling node features to not only include their own state but also incorporate the features of their closely related neighboring nodes. This enhances the sensitivity of feature representation to associated features. Furthermore, by weighted aggregation of the features of neighboring nodes, it effectively suppresses the interference of isolated noise features, improving the stability and robustness of node features. Thus, through spatial aggregation, each node ultimately obtains more complete and expressive updated spatial features.

[0111] (5) Compress the update space features of each node in the time dimension to obtain the spatiotemporal correlation features of the node, and output the matrix composed of the spatiotemporal correlation features of all nodes as the spatiotemporal correlation feature matrix.

[0112] Specifically, step (5) is the time compression process. In practice, for each node, the TCN module can be used to compress its update space features to achieve dimensionality reduction. For details on the working principle of the TCN module, please refer to the description in the relevant technologies, which will not be repeated here.

[0113] Furthermore, the dimension of the spatiotemporal correlation feature is set according to actual needs. In this embodiment, it is not limited. For example, in one possible implementation, the dimension of the spatiotemporal correlation feature can be half the dimension of the updated spatial feature.

[0114] Continuing with the example above, for instance, for each node, the dimension of its spatiotemporal correlation features is N. 1. In this embodiment, through compression, the dimension of the spatiotemporal correlation features can be reduced to N / 2. 1.

[0115] Furthermore, the spatiotemporal correlation features of all nodes are combined to form a spatiotemporal correlation feature matrix. Using the example above, this spatiotemporal correlation feature matrix has a dimension of N / 2. 12.

[0116] As described above, it is understandable that spatial aggregation can integrate the weighted information of neighboring nodes in the spatial dimension, highlighting the associations that are identifiable for fault identification. At the same time, by compressing the time dimension, redundant time-series data can be effectively reduced, the expressive power of features can be effectively improved, and the accuracy and reliability of fault identification can be further enhanced.

[0117] Furthermore, in combination Figure 4 After obtaining the spatiotemporal correlation feature matrix, the classification module identifies the spatiotemporal correlation feature matrix.

[0118] Understandably, the classification module can be a typical classifier, capable of classifying the spatiotemporal correlation feature matrix, determining the probability of it belonging to each preset category, and then outputting the target category corresponding to the highest probability as the fault identification result. Specifically, the implementation principle and process of the classification module classifying the spatiotemporal correlation feature matrix can be found in relevant technical descriptions, and will not be elaborated upon here.

[0119] The fault detection method provided in this embodiment identifies faults in a relational graph. First, in the spatial dimension, it captures local dependencies between nodes by weighted aggregation of neighbor node features, enhancing the sensitivity of feature representation to associated information and suppressing the interference of isolated noise. Second, in the temporal dimension, it extracts dynamic evolution patterns across time by compressing features that change over time, reducing redundant information and improving the compactness and robustness of feature representation. Thus, by combining spatial aggregation and temporal compression, the identification model can generate a spatiotemporal correlation feature matrix that reflects both spatial dependencies and temporal evolution. Furthermore, using the spatiotemporal correlation feature matrix for fault identification can further improve the accuracy and stability of fault identification.

[0120] Figure 5 This is a flowchart of Embodiment 3 of the fault detection method provided in this application. Please refer to... Figure 5 Based on the above embodiments, the fault detection method provided in this embodiment further includes:

[0121] S501. Receive the sound data blocks collected in real time by the acquisition device during the operation of the device under test.

[0122] Specifically, in one possible implementation, the acquisition device is used to acquire acoustic data during the operation of the device under test in real time, and when the amount of acquired acoustic data reaches a preset threshold, the currently acquired acoustic data is sent as a sound data block to the fault detection device.

[0123] It should be noted that the preset threshold is set according to actual needs, and is not limited in this embodiment. For example, the preset threshold can be 10KB, 20KB, etc. The following explanation takes a preset threshold of 10KB as an example. In this case, the acquisition device will send 10KB of acoustic data as a sound data packet to the fault detection device every time it acquires 10KB of acoustic data.

[0124] Understandably, after acquiring analog signals, the acquisition device can first convert them into digital signals and then send them to the fault detection device.

[0125] S502. Determine whether a specified number of sound data blocks have been stored between the first storage area pointed to by the head pointer and the second storage area pointed to by the tail pointer of the pre-constructed circular storage queue. If yes, proceed to step S503; otherwise, proceed to step S505.

[0126] The circular storage queue includes multiple storage areas; each storage area is used to store one block of audio data.

[0127] S503. Combine the specified number of sound data blocks into a sound frame and store the sound frame in the storage area.

[0128] S504. Delete the acoustic data block currently stored in the first storage area, and update the position of the queue head pointer to the next storage area after the first storage area.

[0129] S505. Store the sound data block in the second storage area, and update the position of the tail pointer to the next storage area after the second storage.

[0130] Specifically, Figure 6 This is a schematic diagram illustrating a circular storage queue as an exemplary embodiment of this application. Wherein, Figure 6 Figure (A) in the diagram is a schematic diagram of the initial state of the circular storage queue. Figure 6 Figures (B), (C), and (D) in the diagram are schematic diagrams of the intermediate states of a circular storage queue. The initial state refers to the state when no data is stored, and the intermediate state refers to the state when data has been stored.

[0131] Please refer to Figure 6 The circular storage queue includes multiple storage areas 61, which together form a circular storage space. This circular buffer space is used to store audio data blocks from the acquisition device; specifically, each storage area 61 stores one audio data block.

[0132] It should be noted that the number of storage areas included in the circular storage queue is set according to actual needs, and is not limited in this embodiment. For example, in one possible implementation, the circular storage queue may include 10 to 15 storage areas 61.

[0133] In addition, the amount of data that each storage area can store is the same as the size of a single audio data block from the acquisition device.

[0134] For further details, please refer to [link / reference]. Figure 6 The circular storage queue is equipped with a head pointer 62 and a tail pointer 63. The head pointer 62 is used to indicate the beginning position of the queue and the storage area where the currently deletable data block is located; while the tail pointer 63 is used to indicate the end position of the queue and the storage area where the currently writable data block is located.

[0135] Please refer to Figure 6 In Figure (A), in the initial state, the circular storage queue has not yet stored any data. At this time, both the head pointer 62 and the tail pointer 63 point to the first storage area (pointing to storage area 1).

[0136] Furthermore, after receiving a sound data block, in step S502, it is first determined whether the storage area between the first storage area pointed to by the head pointer of the circular storage queue and the second storage area pointed to by the tail pointer has already stored a specified number of sound data blocks.

[0137] Specifically, the storage area between the first storage area pointed to by the head pointer and the second storage area pointed to by the tail pointer refers to the storage area between the first storage area and the first storage area along the clockwise direction.

[0138] Furthermore, the specific value of the specified quantity is set according to actual needs. In this embodiment, it is not limited, as long as the specified quantity is less than the number of storage areas contained in the circular storage queue. For example, when the circular storage queue contains M storage areas, the specified quantity can be M-1, M-2, etc. The following explanation uses the specified quantity of M-1 as an example.

[0139] Combination Figure 6 In Figure (A), for example, in the initial state, if a sound data block is received, in step S502, it is determined that there are no specified number of sound data blocks stored between the first and second storage areas. Then, step S503 is executed, storing the data block in the second storage area pointed to by the tail pointer, and updating the tail pointer to the next position after the second storage area. After the tail pointer is updated, the process proceeds as follows: Figure 6 The intermediate state shown in Figure (B) is as follows.

[0140] It should be noted that "next position" refers to the next position in a clockwise direction. (Combined with...) Figure 6 In diagram (A), the tail pointer is updated to the position corresponding to storage area 2.

[0141] For example, in one embodiment, such as Figure 6 As shown in Figure (B), at this time, another audio data block is received. In step S502, it is determined that there are still no specified number of audio data blocks stored between the first storage area and the second storage area. At this time, step S505 is executed to store the data block in the second storage area pointed to by the tail pointer and update the tail pointer to the next position.

[0142] For example, in one embodiment, such as Figure 6 As shown in Figure (C), at this time, another sound data block is received. In step S502, it is determined that a specified number of sound data blocks (M-1 sound data blocks have been stored) have been stored between the first storage area and the second storage area. At this time, step S503 is executed. First, the specified number of sound data blocks between the first storage area and the second storage area are combined into a sound frame, and the sound frame is stored in the storage area.

[0143] Specifically, a specified number of audio data blocks between the first and second storage areas are concatenated in chronological order to obtain an audio frame. It can be understood that an audio frame is composed of M-1 audio data blocks arranged in chronological order.

[0144] Furthermore, after executing step S503, steps S504 and S505 are executed sequentially. Specifically, in step S504, the acoustic data block currently stored in the first storage area is deleted, and the position of the head pointer is updated to the next storage area after the first storage area; in step S505, the sound data block is stored in the second storage area, and the position of the tail pointer is updated to the next storage area after the second storage area. After executing S504 and S505, the state of the circular storage queue is as follows. Figure 6 As shown in Figure (D).

[0145] For further details, please refer to [link / reference]. Figure 6In Figure (D), if a sound data block is received, in step S502, it is determined that M-1 acoustic data blocks are stored between the first storage area (storage area 2) pointed to by the head pointer and the second storage area (storage area 1) pointed to by the tail pointer. Then, step S503 is executed, storing the M-1 acoustic data blocks between storage area 2 and storage area 1 as a single sound frame (M-1 sound data blocks arranged clockwise). Further, step S504 is executed, deleting the data block stored in storage area 2 pointed to by the head pointer and updating the position of the head pointer to the corresponding position in storage area 3. Further, step S505 is executed, storing the newly arrived acoustic data block in storage area 1 pointed to by the tail pointer and updating the tail pointer to the corresponding position in storage area 2.

[0146] In this way, when a new data block is received, the M-1 acoustic data blocks between storage area 3 and storage area 2 can be stored as a sound frame, and so on in a loop.

[0147] It is understandable that for any two audio frames, only one audio data block differs, while the other audio data blocks are the same. Following the storage method described above, the first audio data block to the (M-1)th audio data block constitutes one audio frame, the second audio data block to the Mth audio data block constitutes another, the third audio data block to the (M+1)th audio data block constitutes yet another, and so on, in chronological order. This allows for rapid overlapping and framing, resulting in multiple audio frames.

[0148] The fault detection method provided in this embodiment manages continuously acquired acoustic data blocks using a circular storage queue. When the number of acoustic data blocks reaches a specified number, they can be immediately combined into a sound frame, achieving real-time and fast frame segmentation. This ensures that each newly acquired sound data block can be quickly incorporated into the processing flow, avoiding delays caused by waiting or batch processing. As a result, efficient and continuous frame segmentation based on real-time signals is achieved, providing timely and stable input for subsequent fault detection.

[0149] Furthermore, by using a circular storage queue, overlapping framing can be achieved. Firstly, by retaining a certain proportion of common data between adjacent frames, acoustic data from the same time period can be repeatedly presented in multiple frames, thereby increasing the probability of capturing anomalous features and avoiding the loss of boundary features. Secondly, overlapping framing also makes the sound frames smoother and more continuous, which helps the recognition model learn sound change trends more accurately, thus achieving more sensitive detection of weak and transient fault signals and significantly improving the stability and robustness of fault identification.

[0150] Optionally, in one possible implementation, in Figure 5Based on the illustrated embodiment, acquiring acoustic data during the operation of the device under test includes:

[0151] When the number of sound frames stored in the storage area reaches a preset number of frames, the preset number of sound frames are retrieved from the storage area and used as acoustic data during the operation of the device under test.

[0152] In a specific implementation, in one possible approach, when the number of audio frames stored in the storage area is greater than or equal to a preset number of frames, the preset number of audio frames are retrieved from the storage area in chronological order (from back to front) for fault detection. After retrieving the preset number of audio frames from the storage area, these preset number of frames can be deleted or retained; this embodiment does not impose any limitations on this.

[0153] The fault detection method provided in this embodiment accumulates a preset number of sound frames in the storage area and then uses these sound frames as acoustic data of the device under test for fault detection. This allows for fault detection based on the latest real-time signals, which not only improves the timeliness of data processing but also stably manages continuously acquired acoustic data, avoiding delays caused by data loss or excessive accumulation. This facilitates rapid and accurate detection of the operating status of the device under test and improves the response speed and reliability of fault detection.

[0154] Corresponding to the aforementioned embodiment of a fault detection method, this application also provides an embodiment of a fault detection device.

[0155] An embodiment of a fault detection device disclosed in this application can be applied to a fault detection device. The device embodiment can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of the fault detection device loading the corresponding computer program instructions from non-volatile memory into memory and executing them. From a hardware perspective, such as... Figure 7 As shown, Figure 7 This is a hardware structure diagram of a fault detection device within a fault detection equipment, as shown in an exemplary embodiment of this application. (Except for...) Figure 7 In addition to the processor, memory, network interface, and non-volatile memory shown, the fault detection device in the embodiment may also include other hardware depending on the actual function of the fault detection device, which will not be described in detail here.

[0156] Figure 8 This is a schematic diagram of the structure of Embodiment 1 of the fault detection device provided in this application. Please refer to... Figure 8 The fault detection device provided in this embodiment includes an acquisition module 810, an extraction module 820, a construction module 830, and a recognition model 840; wherein,

[0157] The acquisition module 810 is used to acquire acoustic data during the operation of the device under test;

[0158] The extraction module 820 is used to extract features from the acoustic data to obtain a global feature matrix corresponding to the acoustic data; wherein, the global feature matrix includes multi-dimensional features;

[0159] The construction module 830 is used to construct a relational graph corresponding to the global feature matrix; wherein, the relational graph uses each feature dimension in the global feature matrix as a node, the directed connection relationship between any two features as an edge, and the correlation coefficient between the two features as the weight of the edge;

[0160] The identification module 840 is used to perform fault identification on the relationship graph using a pre-trained identification model to obtain the fault identification result of the device under test.

[0161] The apparatus of this embodiment can be used to perform... Figure 1 The steps of the method embodiment shown are similar in principle and process, and will not be repeated here.

[0162] Optionally, in one possible implementation, the building module 830 is specifically used for:

[0163] Extract each dimension feature from the global feature matrix, and use each dimension feature as a node to determine the set of all nodes as the node set;

[0164] For any two-dimensional feature in the global feature matrix, construct the directed edges corresponding to the two-dimensional feature, and determine the set of all directed edges as the edge set;

[0165] For any two-dimensional features in the global feature matrix, determine the correlation coefficient between the two-dimensional features.

[0166] The correlation coefficient between any two-dimensional features is determined as the weight of the directed edge corresponding to the two-dimensional features, and the matrix formed by the weights of all directed edges is determined as the weight matrix.

[0167] The combination of the node set, the edge set, and the weight matrix is ​​used to determine the relation graph.

[0168] Optionally, in one possible implementation, the recognition model includes a spatiotemporal attention module and a classification module; the spatiotemporal attention module includes at least one spatiotemporal attention layer; wherein,

[0169] Each spatiotemporal attention layer is used to perform spatial aggregation and temporal compression on the feature matrix currently represented by the node set based on the edge set and the weight matrix, so as to obtain the spatiotemporal correlation feature matrix corresponding to the global feature matrix;

[0170] The classification module is used to perform fault identification on the spatiotemporal correlation feature matrix to obtain the fault identification result.

[0171] Optionally, in one possible implementation, performing spatial aggregation and temporal compression on the feature matrix currently represented by the node set based on the edge set and the weight matrix includes:

[0172] For each node in the node set, determine the set of neighboring nodes corresponding to that node based on the edge set;

[0173] For each edge in the set of edges, the attention coefficient of the edge is determined based on the weight of the edge and the weights of the associated edges; wherein the associated edges are the edges between the starting node of the edge and each of the neighboring nodes of the starting node.

[0174] For each node in the node set, the current features of the node's neighbors are weighted according to the attention coefficients of the edges between the node and its neighbors to obtain the updated feature increment of the node.

[0175] Based on the current characteristics of the node and the update feature increment of the node, determine the update space characteristics of the node;

[0176] The update space features of each node are compressed in the time dimension to obtain the spatiotemporal correlation features of that node, and the matrix formed by the spatiotemporal correlation features of all nodes is output as the spatiotemporal correlation feature matrix.

[0177] Optionally, in one possible implementation, the fault detection method further includes a processing module, which is specifically used for:

[0178] Receives sound data blocks collected in real time by the acquisition device during the operation of the device under test;

[0179] Determine whether a specified number of audio data blocks have been stored between the first storage area pointed to by the head pointer and the second storage area pointed to by the tail pointer of the pre-constructed circular storage queue; wherein, the circular storage queue includes multiple storage areas; each storage area is used to store one audio data block;

[0180] If not, store the audio data block in the second storage area and update the position of the tail pointer to the next storage area after the second storage area;

[0181] If so, combine the specified number of sound data blocks into a sound frame and store the sound frame in the storage area;

[0182] Delete the acoustic data block currently stored in the first storage area, and update the position of the queue head pointer to the next storage area after the first storage area;

[0183] The sound data block is stored in the second storage area, and the position of the tail pointer is updated to the next storage area after the second storage.

[0184] Optionally, in one possible implementation, the acquisition module 810 is specifically used to acquire the preset number of sound frames from the storage area when the number of sound frames stored in the storage area reaches a preset number of frames, and use the acquired preset number of sound frames as acoustic data during the operation of the device under test.

[0185] Optionally, in one possible implementation, the acoustic data includes a preset number of sound frames; the extraction module 820 is specifically used to extract the multidimensional features of each sound frame for each sound frame in the preset number of sound frames, to obtain a local feature vector of each sound frame.

[0186] The extraction module 820 is also specifically used to concatenate the local feature vectors of the preset number of sound frames in chronological order to obtain the global feature matrix of the acoustic data.

[0187] Please continue to refer to Figure 7 This application also provides a fault detection device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of any of the fault detection methods provided in the first aspect of this application.

[0188] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0189] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0190] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A fault detection method, characterized in that, The fault detection method includes: Acquire acoustic data during the operation of the device under test; Feature extraction is performed on the acoustic data to obtain a global feature matrix corresponding to the acoustic data; wherein, the global feature matrix includes multi-dimensional features; Construct a relational graph corresponding to the global feature matrix; wherein, the relational graph uses each feature dimension in the global feature matrix as a node, the directed connection relationship between any two features as an edge, and the correlation coefficient between the two features as the weight of the edge; The fault identification results of the device under test are obtained by using a pre-trained recognition model to perform fault identification on the relationship graph. The recognition model includes a spatiotemporal attention module and a classification module; the spatiotemporal attention module includes at least one spatiotemporal attention layer; each spatiotemporal attention layer is used to determine the set of neighboring nodes corresponding to each node based on the edge. The spatiotemporal attention layer is further configured to determine the attention coefficient of each edge based on the weight of the edge and the weights of its associated edges; wherein the associated edges are the edges between the starting node of the edge and each of its neighboring nodes. The spatiotemporal attention layer is also used to, for each node, perform weighted processing on the current features of the neighboring nodes of the node according to the attention coefficient of the edge between the node and the neighboring nodes of the node, to obtain the update feature increment of the node, and determine the update spatial features of the node according to the current features of the node and the update feature increment of the node. The spatiotemporal attention layer is also used to compress the update spatial features of each node in the time dimension to obtain the spatiotemporal correlation features of the node, and output the matrix composed of the spatiotemporal correlation features of all nodes as the spatiotemporal correlation feature matrix. The classification module is used to perform fault identification on the spatiotemporal correlation feature matrix to obtain the fault identification result.

2. The fault detection method according to claim 1, characterized in that, The construction of the relational graph corresponding to the global feature matrix includes: Extract each dimension feature from the global feature matrix, and use each dimension feature as a node to determine the set of all nodes as the node set; For any two-dimensional feature in the global feature matrix, construct the directed edges corresponding to the two-dimensional feature, and determine the set of all directed edges as the edge set; For any two-dimensional features in the global feature matrix, determine the correlation coefficient between the two-dimensional features. The correlation coefficient between any two-dimensional features is determined as the weight of the directed edge corresponding to the two-dimensional features, and the matrix formed by the weights of all directed edges is determined as the weight matrix. The combination of the node set, the edge set, and the weight matrix is ​​used to determine the relation graph.

3. The fault detection method according to claim 1, characterized in that, The fault detection method further includes: Receives sound data blocks collected in real time by the acquisition device during the operation of the device under test; Determine whether a specified number of audio data blocks have been stored between the first storage area pointed to by the head pointer and the second storage area pointed to by the tail pointer of the pre-constructed circular storage queue; wherein, the circular storage queue includes multiple storage areas; each storage area is used to store one audio data block; If not, store the audio data block in the second storage area and update the position of the tail pointer to the next storage area after the second storage area; If so, combine the specified number of sound data blocks into a sound frame and store the sound frame in the storage area; Delete the acoustic data block currently stored in the first storage area, and update the position of the queue head pointer to the next storage area after the first storage area; The sound data block is stored in the second storage area, and the position of the tail pointer is updated to the next storage area after the second storage.

4. The fault detection method according to claim 3, characterized in that, The acquisition of acoustic data during the operation of the device under test includes: When the number of sound frames stored in the storage area reaches a preset number of frames, the preset number of sound frames are retrieved from the storage area and used as acoustic data during the operation of the device under test.

5. The fault detection method according to claim 1 or claim 4, characterized in that, The acoustic data includes a preset number of sound frames; the feature extraction of the acoustic data to obtain the global feature matrix corresponding to the acoustic data includes: For each sound frame in the preset number of sound frames, the multidimensional features of each sound frame are extracted to obtain the local feature vector of each sound frame. The local feature vectors of the preset number of sound frames are concatenated in chronological order to obtain the global feature matrix of the acoustic data.

6. The fault detection method according to claim 2, characterized in that, Determining the correlation coefficient between any two features in the global feature matrix includes: For each feature dimension in the global feature matrix, each element in that feature dimension is normalized to obtain the standardized feature matrix corresponding to the global feature matrix. Calculate the covariance matrix of the standardized feature matrix, and determine the Pearson correlation coefficient between any two features based on the covariance matrix; The absolute value of the Pearson correlation coefficient between any two features is determined as the correlation coefficient between the two features.

7. A fault detection device, characterized in that, The fault detection device includes an acquisition module, an extraction module, a construction module, and an identification module; wherein, The acquisition module is used to acquire acoustic data during the operation of the device under test; The extraction module is used to extract features from the acoustic data to obtain a global feature matrix corresponding to the acoustic data; wherein, the global feature matrix includes multi-dimensional features; The construction module is used to construct a relational graph corresponding to the global feature matrix; wherein, the relational graph uses each feature dimension in the global feature matrix as a node, the directed connection relationship between any two features as an edge, and the correlation coefficient between the two features as the weight of the edge; The identification module is used to perform fault identification on the relationship graph using a pre-trained identification model to obtain the fault identification result of the device under test. The recognition model includes a spatiotemporal attention module and a classification module; the spatiotemporal attention module includes at least one spatiotemporal attention layer; each spatiotemporal attention layer is used to determine the set of neighboring nodes corresponding to each node based on the edge. The spatiotemporal attention layer is further configured to determine the attention coefficient of each edge based on the weight of the edge and the weights of its associated edges; wherein the associated edges are the edges between the starting node of the edge and each of its neighboring nodes. The spatiotemporal attention layer is also used to, for each node, perform weighted processing on the current features of the neighboring nodes of the node according to the attention coefficient of the edge between the node and the neighboring nodes of the node, to obtain the update feature increment of the node, and determine the update spatial features of the node according to the current features of the node and the update feature increment of the node. The spatiotemporal attention layer is also used to compress the update spatial features of each node in the time dimension to obtain the spatiotemporal correlation features of the node, and output the matrix composed of the spatiotemporal correlation features of all nodes as the spatiotemporal correlation feature matrix. The classification module is used to perform fault identification on the spatiotemporal correlation feature matrix to obtain the fault identification result.

8. A fault detection 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 computer program, it implements the steps of the fault detection method according to any one of claims 1 to 6.

Images