An apparatus anomaly detection method, device and storage medium

Abstract

The application provides a device anomaly detection method, device and storage medium. The method comprises: constructing a stacked capsule autoencoder according to a plurality of normal sensor data samples; obtaining a plurality of to-be-detected sensor data and calculating feature images corresponding to the plurality of to-be-detected sensor data; inputting the feature images into the stacked capsule autoencoder to obtain reconstruction residuals of the plurality of to-be-detected sensor data; and comparing the reconstruction residuals with a reconstruction residual range of normal sensor data to determine whether a to-be-detected device is abnormal, wherein the data types and quantities of the plurality of to-be-detected sensor data are consistent with those of the plurality of normal data samples. The embodiment of the application can significantly improve the accuracy of detecting whether a to-be-detected device is in an abnormal state by analyzing a plurality of sensor data based on the correlation of the plurality of sensor data and using a stacked capsule autoencoder.

Description

Technical Field

[0001] This application relates to the field of integrated ship support systems, and more specifically to a method, apparatus and storage medium for detecting equipment anomalies based on multi-sensor data. Background Technology

[0002] With the rapid development of technologies such as electronics and communications, sensor networks are now widely used in many fields, including environmental monitoring, traffic monitoring, safety monitoring, and industrial and agricultural production. Sensor data enables real-time measurement of physical quantities, thereby detecting abnormal phenomena that deviate from normal equipment operating conditions. Anomaly detection has become one of the important applications of sensor networks, receiving increasing attention and application. Therefore, using an efficient and accurate anomaly detection method in sensor networks is particularly important for practical applications.

[0003] A major problem with sensor-based anomaly detection equipment is that the objects being detected are generally in normal working condition, and abnormal working conditions rarely occur. Therefore, the sensors collect a lot of normal state data but very little abnormal data. As a result, there are not enough abnormal state data samples to learn and build a model. Consequently, due to the lack of a large amount of standard anomaly detection data, convolutional neural network algorithms cannot be used for learning and training to identify abnormal states.

[0004] Traditional methods for detecting device anomalies based on sensor data construct a stacked autoencoder model from sensor data under normal conditions. The reconstructed residual of the sensor data under test is then calculated using this model, and its accuracy is determined by whether it falls within the range of the normal reconstructed residual. However, traditional anomaly detection based on sensor data only analyzes whether sensor data is abnormal in isolation, ignoring the correlation between different sensor data points. Sometimes, data collected by a sensor may appear normal under one condition but abnormal under another. Traditional methods based on sensor data cannot detect such situations, resulting in low accuracy.

[0005] Therefore, improvements are needed to address the aforementioned issues. Summary of the Invention

[0006] This application is proposed to address the aforementioned problems. According to one aspect of this application, a method for detecting equipment malfunctions is provided, the method comprising:

[0007] A stacked capsule autoencoder was constructed based on data samples from multiple normal sensors.

[0008] Acquire data from multiple sensors to be detected, and calculate the feature images corresponding to the multiple sensor data to be detected;

[0009] The feature image is input into the stacked capsule autoencoder to obtain the reconstruction residual of the multiple sensor data to be detected.

[0010] The reconstructed residual is compared with the reconstructed residual range of normal sensor data to determine whether the device under test is abnormal.

[0011] The data types of the multiple sensor data to be detected are consistent with the data types and quantities of the multiple normal data samples.

[0012] In one embodiment of this application, acquiring multiple sensor data to be detected and calculating the feature image corresponding to the sensor data to be detected includes:

[0013] Acquire data from the plurality of sensors to be detected;

[0014] Each sensor data to be detected is subjected to variational mode decomposition to obtain multiple modal components corresponding to each sensor data to be detected.

[0015] Each modal component is then subjected to variational mode decomposition again to obtain a feature set for each modal component;

[0016] Based on the data from the multiple sensors to be detected, the number of modal components obtained by variational mode decomposition of each sensor, and the number of features of each modal component, a feature image of the sensor data to be detected is constructed.

[0017] In one embodiment of this application, variational mode decomposition is performed on each of the sensor data to be detected to obtain multiple modal components corresponding to each sensor data to be detected, including:

[0018] Each sensor data to be detected is decomposed into a preset number of modal components;

[0019] The Hilbert transform is used to convert the preset number of modal components from real signals into analytic signals;

[0020] Construct a constrained variational model, wherein the optimization function of the constrained variational model is:

[0021]

[0022] Where x(t) represents the signal to be decomposed; x k (t) represents the modal components, 1≤k≤K; K represents the number of pre-decomposed modal components, K is a natural number greater than 1; ω kδ(t) represents the center frequency of each analytical signal; δ(t) represents the pulse signal. It is the partial derivative with respect to time t;

[0023] In the Fourier time and frequency domains, the optimization function is optimized using an alternating direction multiplication algorithm until the error reaches a preset error range.

[0024] In one embodiment of this application, each modal component is subjected to variational mode decomposition again to obtain a feature set for each modal component, including:

[0025] Each modal component is then subjected to variational mode decomposition again to obtain multiple features of each modal component;

[0026] A preset number of features are arbitrarily selected from the plurality of features to form the feature group of the modal component;

[0027] The calculation formula for the plurality of features is as follows:

[0028]

[0029]

[0030]

[0031] T4 k =|max(x) k [n])-min(x k [n])| (5),

[0032]

[0033]

[0034]

[0035] T8 k =-T3 k log(T3 k (9),

[0036] Where, x k [n] represents the modal components, T1 k T2 represents the mean. k T3 represents the root mean square. k x represents k The energy ratio of [n], T4 k T5 represents the absolute peak-to-peak value. k T6 represents variance. k T7 represents the standard deviation. k T8 represents energy.k It represents energy entropy.

[0037] In one embodiment of this application, the preset quantity is greater than or equal to 3 and less than or equal to 8.

[0038] In one embodiment of this application, the reconstruction residual range of normal sensor data is calculated based on the plurality of normal data samples.

[0039] In one embodiment of this application, comparing the reconstructed residual with the reconstructed residual range of normal sensor data to determine whether the device under test is abnormal includes:

[0040] The reconstructed residual is compared with the range of the reconstructed residual of normal sensor data;

[0041] When the reconstruction residual is within the range of the reconstruction residual of the normal sensor data, the device to be tested corresponding to the sensor data to be tested is determined to be in a normal state.

[0042] Otherwise, the device to be tested corresponding to the sensor data to be tested is in an abnormal state.

[0043] According to another aspect of this application, a device for detecting equipment malfunctions is provided, the device comprising:

[0044] Build modules are used to construct stacked capsule autoencoders based on multiple normal sensor data samples;

[0045] The acquisition module is used to acquire data from multiple sensors to be detected and to calculate the feature images corresponding to the multiple sensor data to be detected.

[0046] The input module is used to input the feature image into the stacked capsule autoencoder to obtain the reconstruction residual of the multiple sensor data to be detected;

[0047] The comparison module is used to compare the reconstructed residual with the reconstructed residual range of normal sensor data to determine whether the device under test is abnormal.

[0048] The data types of the multiple sensor data to be detected are consistent with the data types and quantities of the multiple normal data samples.

[0049] In one embodiment of this application, the acquisition module includes:

[0050] The first acquisition submodule is used to acquire data from the plurality of sensors to be detected;

[0051] The first decomposition submodule is used to perform variational mode decomposition on each of the sensor data to be detected to obtain multiple modal components corresponding to each of the sensor data to be detected.

[0052] The second decomposition submodule is used to perform variational mode decomposition on each modal component again to obtain the feature set of each modal component;

[0053] The first construction submodule is used to construct a feature image of the sensor data to be detected based on the data of the plurality of sensors to be detected, the number of modal components obtained by variational mode decomposition of each sensor, and the number of features of each modal component.

[0054] In one embodiment of this application, the first decomposition submodule is further configured to:

[0055] Each sensor data to be detected is decomposed into a preset number of modal components;

[0056] The Hilbert transform is used to convert the preset number of modal components from real signals into analytic signals;

[0057] Construct a constrained variational model with the following optimization function:

[0058]

[0059] Where x(t) represents the signal to be decomposed; x k (t) represents the modal components, 1≤k≤K; K represents the number of pre-decomposed modal components, K is a natural number greater than 1; ω k δ(t) represents the center frequency of each analytical signal; δ(t) represents the pulse signal. It is the partial derivative with respect to time t;

[0060] In the Fourier time and frequency domains, the optimization function is optimized using an alternating direction multiplication algorithm until the error reaches a preset error range.

[0061] In one embodiment of this application, the second decomposition submodule is further configured to:

[0062] Each modal component is then subjected to variational mode decomposition again to obtain multiple features of each modal component;

[0063] A preset number of features are arbitrarily selected from the plurality of features to form the feature group of the modal component;

[0064] The calculation formula for the plurality of features is as follows:

[0065]

[0066]

[0067]

[0068] T4 k =|max(x) k [n])-min(x k [n])| (5),

[0069]

[0070]

[0071]

[0072] T8 k =-T3 k log(T3 k (9),

[0073] Where, x k [n] represents the modal components, T1 k T2 represents the mean. k T3 represents the root mean square. k x represents k The energy ratio of [n], T4 k T5 represents the absolute peak-to-peak value. k T6 represents variance. k T7 represents the standard deviation. k T8 represents energy. k It represents energy entropy.

[0074] In one embodiment of this application, the comparison module is further configured to:

[0075] The reconstructed residual is compared with the range of the reconstructed residual of normal sensor data;

[0076] When the reconstruction residual is within the range of the reconstruction residual of the normal sensor data, the device to be tested corresponding to the sensor data to be tested is determined to be in a normal state.

[0077] Otherwise, the device to be tested corresponding to the sensor data to be tested is in an abnormal state.

[0078] According to another aspect of this application, a storage medium is provided, on which a computer program is stored, which, when run by a processor, causes the processor to execute the above-described device anomaly detection method.

[0079] According to the device anomaly detection method, apparatus and storage medium of this application, based on the correlation of multiple sensor data, a stacked capsule self-encoder is used to analyze multiple sensor data, which can significantly improve the accuracy of detecting whether the device under test is in an abnormal state. Attached Figure Description

[0080] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0081] Figure 1 A schematic flowchart illustrating a device anomaly detection method according to an embodiment of this application is shown;

[0082] Figure 2 A schematic flowchart illustrating the construction of a stacked capsule autoencoder according to an embodiment of this application is shown;

[0083] Figure 3 A schematic flowchart illustrating the calculation of feature images of the normal data sample according to embodiments of this application is shown.

[0084] Figure 4 A schematic flowchart illustrating variational mode decomposition of each sensor data to be detected according to an embodiment of this application is shown.

[0085] Figure 5 A schematic flowchart illustrating the variational mode decomposition of each modal component according to an embodiment of this application is shown.

[0086] Figure 6 A schematic diagram showing one of the modal components s according to an embodiment of this application is illustrated;

[0087] Figure 7 A schematic diagram illustrating mode decomposition of signal s according to an embodiment of this application is shown;

[0088] Figure 8(a) shows a schematic diagram of the data collected by the sensor 1 and sensor 2 under normal conditions according to the embodiments of this application;

[0089] Figure 8(b) shows a schematic diagram of data collected by the sensor 1 and sensor 2 under abnormal conditions according to an embodiment of the present application;

[0090] Figure 9 A schematic block diagram of a device anomaly detection apparatus according to an embodiment of this application is shown. Detailed Implementation

[0091] To make the objectives, technical solutions, and advantages of this application more apparent, exemplary embodiments according to this application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein. Based on the embodiments of this application described herein, all other embodiments obtained by those skilled in the art without inventive effort should fall within the protection scope of this application.

[0092] To address the aforementioned technical problems, this application provides a method for detecting device anomalies, comprising: constructing a stacked capsule autoencoder based on multiple normal sensor data samples; acquiring multiple sensor data to be detected and calculating feature images corresponding to the multiple sensor data to be detected; inputting the feature images into the stacked capsule autoencoder to obtain reconstruction residuals of the multiple sensor data to be detected; comparing the reconstruction residuals with the reconstruction residual range of normal sensor data to determine whether the device to be detected is abnormal; wherein, the data type of the multiple sensor data to be detected is consistent with the data type and quantity of the multiple normal data samples. This embodiment of the invention, based on the correlation of multiple sensor data, uses a stacked capsule autoencoder to analyze multiple sensor data, which can significantly improve the accuracy of detecting whether the device to be detected is in an abnormal state.

[0093] The device anomaly detection method according to embodiments of this application will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, features of the various embodiments of this application can be combined with each other.

[0094] Figure 1 This diagram illustrates a schematic flowchart of a device anomaly detection method according to an embodiment of this application; as shown. Figure 1 As shown, the device anomaly detection method 100 according to an embodiment of this application may include the following steps S101, S102 and S103.

[0095] In step S101, a stacked capsule autoencoder is constructed based on multiple normal sensor data samples.

[0096] In one example, such as Figure 2 As shown, constructing a stacked capsule autoencoder based on multiple normal sensor data samples includes the following steps: A, acquiring multiple normal data samples, wherein each normal data sample includes a set of sensor data; B, calculating the feature image of each normal data sample; C, training the stacked capsule autoencoder based on the feature image of each normal data sample.

[0097] like Figure 3 The diagram shown is a schematic flowchart for calculating the feature image of the normal data sample. Step B includes:

[0098] B1, Perform Variational Mode Decomposition (VMD) on each sensor data in each normal data sample to obtain multiple modal components;

[0099] B2, perform variational mode decomposition on each modal component again to obtain multiple features of each modal component;

[0100] B3. Based on the number of sensors in each normal data sample, the number of modal components obtained by variational mode decomposition of each sensor, and the number of features of each modal component, a feature image of each data sample is obtained.

[0101] In one example, when constructing the stacked capsule autoencoder, the dimension of its input layer is consistent with the dimension of the feature image of each data sample, and the dimension of its output layer is consistent with the dimension of the input layer.

[0102] In a specific embodiment, it is assumed that there are a total of M data samples, each data sample includes a set of sensor data, and each sensor data is complete and normal. A stacked capsule autoencoder can be constructed based on the M data samples.

[0103] Specifically, each sensor data point in each data sample is a time signal x(t). This signal is first subjected to variational mode decomposition, which decomposes it into K modal components x. k (t), correspondingly, for the sample, the sampled value is x. k [n]; where the process of performing variational mode decomposition of the signal is as follows:

[0104] The first step is to decompose each signal x(t) into a predetermined number of modal components, for example, into K modal components, each modal component being x. k (t), and then the Hilbert transform is used to extract the preset number of modal components from the real signal x. k (t) is converted into an analytic signal s k (t).

[0105] The second step is to estimate the analytical signal s for each mode. k The center frequency ω of (t) k By multiplying by e -jωkt The corresponding spectrum is modulated onto the baseband to achieve frequency mixing. The L2 norm of the demodulated signal gradient is calculated, and the sum of the bandwidths of each mode is minimized as a constraint. The bandwidth of each mode is estimated, and a constrained variational model is constructed. Its optimization function is:

[0106]

[0107] Where x(t) represents the signal to be decomposed; x k (t) represents the modal components, 1≤k≤K; K represents the number of pre-decomposed modal components, K is a natural number greater than 1; ω k δ(t) represents the center frequency of each analytical signal; δ(t) represents the pulse signal. It is the partial derivative with respect to time t;

[0108] The third step involves optimizing the function from the second step using an alternating direction multiplication algorithm in both the Fourier time and frequency domains until the error reaches a preset error range, at which point the iteration stops. The result obtained through the iteration is then subjected to an inverse Fourier transform; its real part is the modal component u in the time domain. k (t).

[0109] Then, after performing variational mode decomposition on the data from each sensor, the modal component x is calculated. k The characteristics of [n] are T1, T2, ..., T8. The formulas for calculating T1 to T8 are as follows:

[0110]

[0111]

[0112]

[0113] T4 k =|max(x) k [n])-min(x k [n])| (5),

[0114]

[0115]

[0116]

[0117] T8 k =-T3 k log(T3 k (9),

[0118] Where, x k [n] represents the modal components, T1 k T2 represents the mean. k T3 represents the root mean square. k x represents k The energy ratio of [n], T4 k T5 represents the absolute peak-to-peak value. k T6 represents variance. kT7 represents the standard deviation. k T8 represents energy. k It represents energy entropy.

[0119] Arbitrarily select three or more features from the above eight features to form a feature group. Let the feature group have L features, where L ranges from {3, 4, ..., 8}. In each sensor data sample, the K modal components of each sensor form a feature vector, which has L x K features. The K modal components of the S sensors in each sample can form a two-dimensional feature matrix with S x (L x K) features. This two-dimensional feature matrix can be viewed as a two-dimensional grayscale image. Copying this two-dimensional grayscale image to the R, G, and B channels creates a three-dimensional color image. The dimension of this three-dimensional color image is S x (L x K) x 3, and the values ​​of the R, G, and B channels are the same. This three-dimensional color image is called the feature image of a sample.

[0120] In step C, the aforementioned feature images can be used as input to train a stacked capsule autoencoder. The dimension of the input layer of this stacked capsule autoencoder corresponds to the dimension of a sample's feature image, and the dimension of its output layer is the same as the dimension of the input layer. The remaining parts adopt the network structure of a standard stacked capsule autoencoder.

[0121] In step S102, the sensor data to be detected is acquired, and the feature image corresponding to the sensor data to be detected is calculated.

[0122] It is worth noting that the data type of the sensor data to be tested should be consistent with the data type of the normal sensor data sample. In this embodiment of the invention, the accuracy of the analysis of the sensor data to be tested can only be ensured when the data type of the sensor data to be tested is consistent with the data type of the normal sensor data sample.

[0123] In one example, the steps of acquiring sensor data to be detected and calculating the feature image corresponding to the sensor data to be detected include: D, acquiring the plurality of sensor data to be detected; E, performing variational mode decomposition on each of the sensor data to be detected to obtain multiple modal components corresponding to each of the sensor data to be detected; F, performing variational mode decomposition on each modal component again to obtain feature groups for each modal component; G, constructing the feature image of the sensor data to be detected based on the data of the plurality of sensor data to be detected, the number of modal components obtained by variational mode decomposition of each sensor, and the number of features of each modal component.

[0124] In step E, the variational mode decomposition of each sensor data to be detected is performed, a process essentially consistent with step B described above. Specifically, as follows... Figure 4As shown, in this embodiment, the steps for performing variational mode decomposition on each of the sensor data to be detected are as follows:

[0125] E1, decompose each sensor data to be detected into a preset number of modal components;

[0126] E2, using Hilbert transform to convert the preset number of modal components from real signals into analytic signals;

[0127] E3, construct a constrained variational model with the following optimization function:

[0128]

[0129] Where x(t) represents the signal to be decomposed; x k (t) represents the modal components, 1≤k≤K; K represents the number of pre-decomposed modal components, K is a natural number greater than 1; ω k δ(t) represents the center frequency of each analytical signal; δ(t) represents the pulse signal. It is the partial derivative with respect to time t;

[0130] In the Fourier time and frequency domains, the optimization function is optimized using an alternating direction multiplication algorithm until the error reaches a preset error range.

[0131] In step F, as Figure 5 As shown, the process of performing variational mode decomposition on each modal component again to obtain the feature set of each modal component is as follows:

[0132] F1, perform variational mode decomposition on each modal component again to obtain multiple features of each modal component;

[0133] F2, arbitrarily select a preset number of features from the plurality of features to form the feature group of the modal component.

[0134] The calculation formulas for the multiple features are as follows:

[0135]

[0136]

[0137]

[0138] T4 k =|max(x) k [n])-min(x k [n])| (5),

[0139]

[0140]

[0141]

[0142] T8 k =-T3 k log(T3 k (9),

[0143] Where, x k [n] represents the modal components, T1 k T2 represents the mean. k T3 represents the root mean square. k x represents k The energy ratio of [n], T4 k T5 represents the absolute peak-to-peak value. k T6 represents variance. k T7 represents the standard deviation. k T8 represents energy. k It represents energy entropy.

[0144] In a specific embodiment, three or more features from T1 to T8 can be arbitrarily selected to form a feature group, and the number of features in each feature group is greater than or equal to 3 and less than or equal to 8.

[0145] In step S103, the feature image is input into the stacked capsule autoencoder to obtain the reconstruction residual corresponding to the sensor data to be detected. If the reconstruction residual is within the range of the reconstruction residual, the device to be detected corresponding to the sensor data to be detected is determined to be in normal condition.

[0146] In a specific example, the reconstruction residual range of normal sensor data is calculated based on the plurality of normal data samples. Specifically, the plurality of normal data samples are first input into the constructed stacked capsule autoencoder, then the reconstruction residual of each normal data sample is calculated, and then the reconstruction residual range of the normal data sample is determined based on the maximum and minimum values ​​of the reconstruction residuals of each normal data sample.

[0147] For example, M data samples are input into a pre-constructed stacked capsule autoencoder, and the reconstruction residual of each sample is recorded. Let the minimum reconstruction residual be Emin and the maximum reconstruction residual be Emax. Then, the reconstruction residual interval obtained from normal sensor data samples is [Emin, Emax]. In this embodiment of the invention, a stacked capsule autoencoder is constructed based on normal sensor data samples. The reconstruction residual range determined based on these normal data samples is the reconstruction residual range corresponding to normal sensor data. Therefore, when the sensor data to be detected is normal sensor data, its reconstruction residual should be within the reconstruction residual range of the normal sensor data samples. If it is not within the reconstruction residual range of the normal sensor data samples, it indicates that the sensor data to be detected is abnormal, and the device may also be abnormal.

[0148] In one specific embodiment of the present invention, six sensors are installed at different parts of the engine of a certain device. Data from these six sensors, such as vibration signals, is collected, and 100 sensor data samples in normal state are obtained. Simultaneously, data from five sensors (sensor 1 and sensor 2) to be tested are collected, with each sample lasting 2 seconds. According to the method of this embodiment, a stacked capsule self-encoder is first constructed based on the 100 sensor data samples in normal state. Then, the residual range of the normal data is obtained by inputting the 100 sensor data samples in normal state. Finally, abnormal data is input for verification. Specifically, each normal state sensor data sample x... k (t) Perform mode decomposition. For example... Figure 6 As shown, this is one of the modal components s. Figure 7 The diagram illustrates the modal decomposition of signal s to obtain three modal components when K=3. Then, variational modal decomposition is performed on each modal component to obtain a feature set and a feature image of 6x(3x8)x3 dimensions. A stacked capsule autoencoder is constructed based on these feature images. The initial learning rate is set to 1e-2, the Adam optimization method is used, the MSE function is employed as the loss function, the batch size is set to 64, and the remaining parameters are left as default. After 50,000 training iterations, the final loss is less than 0.68. Based on the residual calculation results of all normal states, the residual interval for the normal state data is set to [0.0183, 0.0675] according to the maximum and minimum values. Then, the feature images corresponding to the five data samples in the abnormal state are input into the stacked capsule autoencoder.

[0149] Figure 8(a) shows a schematic diagram of the data collected by the sensor 1 and sensor 2 under test when they are normal sensor data, where s1 is the signal collected by sensor 1 and s2 is the signal collected by sensor 2.

[0150] Figure 8(b) illustrates a scenario where the data collected by sensor 1 and sensor 2 are abnormal sensor data, where s1 is the signal collected by sensor 1 and s2 is the signal collected by sensor 2. In this case, the reconstruction residual calculated based on the data collected by sensor 1 and sensor 2 is outside the aforementioned interval and can be clearly detected.

[0151] As can be seen, by utilizing the correlation between data from multiple sensors, the detection of data anomalies is highly accurate through the embodiments of the present invention.

[0152] The following is combined with Figure 9 The device for detecting equipment malfunctions in this application is described. Among them, Figure 9 A schematic block diagram of a device anomaly detection apparatus 900 according to an embodiment of this application is shown. Each module can respectively perform the functions described above in conjunction with... Figure 1 This describes the steps / functions of the equipment anomaly detection method. The following only describes the main functions of each unit of the equipment anomaly detection device 900, omitting the details already described above. For example... Figure 9 As shown, the equipment anomaly detection device 900 includes: a construction module 901, an acquisition module 902, an input module 903, and a comparison module 904.

[0153] Module 301 is used to build a stacked capsule autoencoder based on multiple normal sensor data samples;

[0154] The acquisition module 902 is used to acquire data from multiple sensors to be detected and to calculate the feature images corresponding to the multiple sensor data to be detected.

[0155] Input module 903 is used to input the feature image into the stacked capsule autoencoder to obtain the reconstruction residual of the multiple sensor data to be detected;

[0156] The comparison module 904 is used to compare the reconstructed residual with the reconstructed residual range of normal sensor data to determine whether the device under test is abnormal.

[0157] The data types of the multiple sensor data to be detected are consistent with the data types and quantities of the multiple normal data samples.

[0158] In one example, the acquisition module 902 includes:

[0159] The acquisition submodule 9021 is used to acquire data from the plurality of sensors to be detected;

[0160] The first decomposition submodule 9022 is used to perform variational mode decomposition on each of the sensor data to be detected to obtain multiple modal components corresponding to each of the sensor data to be detected.

[0161] The second decomposition submodule 9023 is used to perform variational mode decomposition on each modal component again to obtain the feature group of each modal component.

[0162] The construction submodule 9024 is used to construct a feature image of the sensor data to be detected based on the data of the multiple sensors to be detected, the number of modal components obtained by variational mode decomposition of each sensor, and the number of features of each modal component.

[0163] In one example, the first decomposition submodule 9022 is further configured to:

[0164] Each sensor data to be detected is decomposed into a preset number of modal components;

[0165] The Hilbert transform is used to convert the preset number of modal components from real signals into analytic signals;

[0166] Construct a constrained variational model with the following optimization function:

[0167]

[0168] Where x(t) represents the signal to be decomposed; x k (t) represents the modal components, 1≤k≤K; K represents the number of pre-decomposed modal components, K is a natural number greater than 1; ω k δ(t) represents the center frequency of each analytical signal; δ(t) represents the pulse signal. It is the partial derivative with respect to time t;

[0169] In the Fourier time and frequency domains, the optimization function is optimized using an alternating direction multiplication algorithm until the error reaches a preset error range.

[0170] In one example, the second decomposition submodule 9023 is also used for:

[0171] Each modal component is then subjected to variational mode decomposition again to obtain multiple features of each modal component;

[0172] A preset number of features are arbitrarily selected from the plurality of features to form the feature group of the modal component;

[0173] The calculation formula for the plurality of features is as follows:

[0174]

[0175]

[0176]

[0177] T4 k =|max(x) k [n])-min(x k [n])| (5),

[0178]

[0179]

[0180]

[0181] T8 k =-T3 k log(T3 k (9),

[0182] Where, x k [n] represents the modal components, T1 k T2 represents the mean. k T3 represents the root mean square. k x represents k The energy ratio of [n], T4 k T5 represents the absolute peak-to-peak value. k T6 represents variance. k T7 represents the standard deviation. k T8 represents energy. k It represents energy entropy.

[0183] In one example, the comparison module 904 is further used for:

[0184] The reconstructed residual is compared with the range of the reconstructed residual of normal sensor data;

[0185] When the reconstruction residual is within the range of the reconstruction residual of the normal sensor data, the device to be tested corresponding to the sensor data to be tested is determined to be in a normal state.

[0186] Otherwise, the device to be tested corresponding to the sensor data to be tested is in an abnormal state.

[0187] Furthermore, according to embodiments of this application, a storage medium is also provided, on which program instructions are stored. When executed by a computer or processor, these program instructions are used to perform corresponding steps of the device anomaly detection method of this application. The storage medium may, for example, include a memory card of a smartphone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media.

[0188] The device anomaly detection apparatus and storage medium of this application embodiment have the same advantages as the aforementioned device anomaly detection method because they can implement the aforementioned device anomaly detection method.

[0189] Although exemplary embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above exemplary embodiments are merely illustrative and are not intended to limit the scope of this application. Various changes and modifications can be made therein by those skilled in the art without departing from the scope and spirit of this application. All such changes and modifications are intended to be included within the scope of this application as claimed in the appended claims.

[0190] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0191] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed.

[0192] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0193] Similarly, it should be understood that, in order to streamline this application and aid in understanding one or more of the various inventive aspects, features of this application may sometimes be grouped together in a single embodiment, figure, or description thereof in the description of exemplary embodiments of this application. However, this approach should not be construed as reflecting an intention that the claimed application requires more features than are expressly recited in each claim. Rather, as reflected in the corresponding claims, its inventive point lies in solving the corresponding technical problem with features fewer than all features of a single disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of this application.

[0194] Those skilled in the art will understand that, apart from the mutual exclusion of features, all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or apparatus so disclosed can be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0195] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features but not others included in other embodiments, combinations of features from different embodiments are intended to be within the scope of this application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0196] The various component embodiments of this application can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some modules according to the embodiments of this application. This application can also be implemented as an apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such an implementation of this application can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0197] It should be noted that the above embodiments are illustrative of this application and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. This application can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.

[0198] The above description is merely a specific embodiment or illustration of the embodiments of this application. The scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A method for detecting equipment malfunctions, characterized in that, The method includes: A stacked capsule autoencoder is constructed based on multiple normal sensor data samples, wherein the stacked capsule autoencoder is used to model the correlation between multiple sensor data; Multiple sensor data to be detected are acquired, and each sensor data to be detected is subjected to variational mode decomposition to obtain multiple modal components corresponding to each sensor data. Each modal component is then subjected to variational mode decomposition again to obtain feature groups of each modal component. Based on the data from multiple sensors, the number of modal components obtained by variational mode decomposition of each sensor, and the number of features of each modal component, a feature image of the multidimensional structure of the sensor data to be detected is constructed. The feature image of the multidimensional structure is input into the stacked capsule autoencoder to obtain the reconstruction residual of the multiple sensor data to be detected. The reconstructed residual is compared with the reconstructed residual range of normal sensor data to determine whether the device under test is abnormal. The data types of the multiple sensor data to be detected are consistent with the data types of the multiple normal data samples, and the number of the multiple sensor data to be detected is consistent with the number of the multiple normal data samples, so as to ensure that the feature image dimensions of the input stacked capsule autoencoder are matched.

2. The method as described in claim 1, characterized in that, Each of the sensor data to be detected is subjected to variational mode decomposition to obtain multiple modal components corresponding to each sensor data to be detected, including: Each sensor data to be detected is decomposed into a preset number of modal components; The Hilbert transform is used to convert the preset number of modal components from real signals into analytic signals; Construct a constrained variational model, wherein the optimization function of the constrained variational model is: (1)， Where x(t) represents the signal to be decomposed; x k (t) represents the modal components, 1≤k≤K; K represents the number of pre-decomposed modal components, K is a natural number greater than 1; ω k δ(t) represents the center frequency of each analytical signal; δ(t) represents the pulse signal, ∂ t It is the partial derivative with respect to time t; In the Fourier time and frequency domains, the optimization function is optimized using an alternating direction multiplication algorithm until the error reaches a preset error range.

3. The method as described in claim 1, characterized in that, Each modal component is then subjected to variational mode decomposition again to obtain a feature set for each modal component, including: Each modal component is then subjected to variational mode decomposition again to obtain multiple features of each modal component; A preset number of features are arbitrarily selected from the plurality of features to form the feature group of the modal component; The calculation formula for the plurality of features is as follows: （2）， （3）， （4）， （5）， （6）， （7）， （8）， （9）， Where, x k [n] represents the modal components, T1 k T2 represents the mean. k T3 represents the root mean square. k x represents k The energy ratio of [n], T4 k T5 represents the absolute peak-to-peak value. k T6 represents variance. k T7 represents the standard deviation. k T8 represents energy. k It represents energy entropy.

4. The method as described in claim 3, characterized in that, in, The preset quantity is greater than or equal to 3 and less than or equal to 8.

5. The method as described in claim 1, characterized in that, in, The reconstructed residual range of normal sensor data is calculated based on the multiple normal data samples.

6. The method as described in claim 1, characterized in that, The reconstructed residual is compared with the reconstructed residual range of normal sensor data to determine whether the device under test is abnormal, including: The reconstructed residual is compared with the range of the reconstructed residual of normal sensor data; When the reconstruction residual is within the range of the reconstruction residual of the normal sensor data, the device to be tested corresponding to the sensor data to be tested is determined to be in a normal state. Otherwise, the device to be tested corresponding to the sensor data to be tested is in an abnormal state.

7. A device for detecting equipment malfunctions, characterized in that, The device includes: A construction module is used to build a stacked capsule autoencoder based on multiple normal sensor data samples, wherein the stacked capsule autoencoder is used to model the correlation between multiple sensor data; The acquisition module is used to acquire data from multiple sensors to be detected, perform variational mode decomposition on each sensor data to obtain multiple modal components, perform variational mode decomposition on each modal component again to obtain feature groups, and construct a multi-dimensional feature image based on the data from multiple sensors, the number of modal components and the number of features. The input module is used to input the feature image of the multidimensional structure into the stacked capsule autoencoder to obtain the reconstruction residual of the multiple sensor data to be detected; The comparison module is used to compare the reconstructed residual with the reconstructed residual range of normal sensor data to determine whether the device under test is abnormal. The data types of the multiple sensor data to be detected are consistent with the data types of the multiple normal data samples, and the number of the multiple sensor data to be detected is consistent with the number of the multiple normal data samples.

8. The apparatus as claimed in claim 7, characterized in that, The acquisition module includes: The first acquisition submodule is used to acquire data from the plurality of sensors to be detected; The first decomposition submodule is used to perform variational mode decomposition on each of the sensor data to be detected to obtain multiple modal components corresponding to each of the sensor data to be detected. The second decomposition submodule is used to perform variational mode decomposition on each modal component again to obtain the feature set of each modal component; The first construction submodule is used to construct a feature image of the sensor data to be detected based on the data of the plurality of sensors to be detected, the number of modal components obtained by variational mode decomposition of each sensor, and the number of features of each modal component.

9. The apparatus as claimed in claim 8, characterized in that, The first decomposition submodule is also used for: Each sensor data to be detected is decomposed into a preset number of modal components; The Hilbert transform is used to convert the preset number of modal components from real signals into analytic signals; Construct a constrained variational model with the following optimization function: (1) Where x(t) represents the signal to be decomposed; x k (t) represents the modal components, 1≤k≤K; K represents the number of pre-decomposed modal components, K is a natural number greater than 1; ω k δ(t) represents the center frequency of each analytical signal; δ(t) represents the pulse signal, ∂ t It is the partial derivative with respect to time t; In the Fourier time and frequency domains, the optimization function is optimized using an alternating direction multiplication algorithm until the error reaches a preset error range.

10. The apparatus as claimed in claim 8, characterized in that, The second decomposition submodule is also used for: Each modal component is then subjected to variational mode decomposition again to obtain multiple features of each modal component; A preset number of features are arbitrarily selected from the plurality of features to form the feature group of the modal component; The calculation formula for the plurality of features is as follows: （2）， （3）， （4）， （5）， （6）， （7）， （8）， （9）， Where, x k [n] represents the modal components, T1 k T2 represents the mean. k T3 represents the root mean square. k x represents k The energy ratio of [n], T4 k T5 represents the absolute peak-to-peak value. k T6 represents variance. k T7 represents the standard deviation. k T8 represents energy. k It represents energy entropy.

11. The apparatus as claimed in claim 7, characterized in that, The comparison module is also used for: The reconstructed residual is compared with the range of the reconstructed residual of normal sensor data; When the reconstruction residual is within the range of the reconstruction residual of the normal sensor data, the device to be tested corresponding to the sensor data to be tested is determined to be in a normal state. Otherwise, the device to be tested corresponding to the sensor data to be tested is in an abnormal state.

12. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, causes the processor to perform the device anomaly detection method as described in any one of claims 1 to 6.

Images