Method for detecting working state of target equipment

By extracting and processing the electromagnetic signal characteristics of the target device, using the support vector classifier and the AlexNet model, combined with the fusion judgment neural network, the problem of low detection accuracy of the working state of the electronic device is solved, and accurate monitoring of the working state of the target device is achieved.

CN119066484APending Publication Date: 2024-12-0336TH RES INST OF CETC
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202410992012.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

In the prior art, the operating state detection accuracy of electronic equipment is low, and it is impossible to accurately monitor the operating state of the target equipment in real time, especially in the absence of built-in sensors.

Method used

By obtaining the original time domain amplitude signal and the first-order differential signal of the time domain amplitude of the target device, extracting preset features, performing wavelet transformation, combining support vector classifiers and improved AlexNet model, the fusion judgment neural network is used to predict the working state. Finally, accurate working status detection results are obtained.

Benefits of technology

It improves detection accuracy and reliability, can accurately monitor the working status of the target equipment in real time, enhances the multi-dimensional utilization of electromagnetic signals, and improves prediction accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066484A_ABST
    Figure CN119066484A_ABST
Patent Text Reader

Abstract

The invention relates to a method for detecting the working state of target equipment, belongs to the technical field of equipment monitoring, and solves the problem of low detection precision when the working state of electronic equipment is detected in the prior art. The working state detection method comprises the following steps: extracting a target working feature vector when target equipment works; performing continuous wavelet transform on the original time domain amplitude signal to obtain a target working feature image; inputting the target working feature vector into a trained target support vector classifier model to obtain a first target working state prediction result when the target equipment works; inputting the target working feature vector and the target working feature image into a pre-trained target improved AlexNet model to obtain a second target working state prediction result when the target equipment works; and fusing the first target working state prediction result and the second target working state prediction result to obtain a final target working state detection result. Rapid detection of the working state of the target equipment is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of equipment monitoring, and in particular to a method for detecting the working state of a target equipment. Background Art

[0002] When electronic devices are working, there are components that radiate electromagnetic wave signals to the outside. Many electronic devices cannot identify the working status of the electronic devices through built-in sensors and need to be detected from the outside. Typical electronic devices include cables, antennas, etc.

[0003] The detection of the working status of traditional electronic devices requires the use of specific sensors, which are generally built into the working circuit. By detecting indicators such as the working voltage, current or frequency of the working circuit, and setting a judgment threshold, the state change is reported after the signal crosses the judgment threshold. However, many electronic devices do not add sensors during the design process for reasons such as cost saving and design defects, or it is impossible to make changes to the electronic devices, making it impossible to detect the working status of the electronic devices.

[0004] At present, the method of classifying and identifying the working status of equipment based on electromagnetic signals based on machine learning generally directly inputs the original electromagnetic signals into the network model; however, this method has low credibility due to its few features and low detection accuracy. Summary of the invention

[0005] In view of the above analysis, an embodiment of the present invention aims to provide a method for detecting the working status of a target device, so as to solve the problem of low detection accuracy when detecting the working status of an electronic device in the prior art.

[0006] An embodiment of the present invention provides a method for detecting a working state of a target device, the method comprising:

[0007] Acquire the original time domain amplitude signal and the first-order differential signal of the time domain amplitude when the target device is working, extract the preset features of the original time domain amplitude signal and the first-order differential signal of the time domain amplitude respectively, and obtain the target working feature vector when the target device is working; perform continuous wavelet transform on the original time domain amplitude signal to obtain the target working feature image;

[0008] Inputting the target working feature vector into the trained target support vector classifier model to obtain a first target working state prediction result when the target device is working;

[0009] Input the target working feature vector and the target working feature image into the pre-trained target improved AlexNet model to obtain the second target working state prediction result when the target device is working;

[0010] Based on the pre-built fusion decision neural network, the first target working state prediction result and the second target working state prediction result are fused to obtain the final target working state detection result.

[0011] Based on the further improvement of the above working status detection method, the preset features include one or more of the following:

[0012] Maximum value;

[0013] Minimum value;

[0014] Crest factor;

[0015] Zero crossing rate;

[0016] Mean square error.

[0017] Based on the further improvement of the above working status detection method, the kernel function used in the training of the target support vector classifier model is:

[0018]

[0019] Among them, x i 、x j represents any two training samples, σ 1 Represents the hyperparameters of the target support vector classifier model.

[0020] Based on the further improvement of the above working status detection method, the optimization conditions satisfied during the training of the target support vector classifier model are:

[0021]

[0022] Among them, N is the total number of training set samples, α i is the Lagrange multiplier of the i-th sample, y i is the category label value of the i-th sample, and C is the penalty parameter;

[0023] Under the premise of the optimization conditions, the following objective function is solved to obtain the Lagrange multiplier of each sample corresponding to the minimum value of the objective function, and the target support vector classifier model training is completed:

[0024]

[0025] Among them, α j is the Lagrange multiplier of the jth sample, y j is the class label value of the jth sample.

[0026] Based on the further improvement of the above working state detection method, the original time domain amplitude signal is transformed by continuous wavelet transform through the following formula to obtain the target working characteristic image:

[0027]

[0028] Among them, s(t) represents the original time domain amplitude signal, S(a,b) represents the target working characteristic image, a represents the translation position, b represents the scaling factor, ψ(t) represents the mother wavelet, represents complex conjugate.

[0029] Based on the further improvement of the above working status detection method, the following function is adopted as the mother wavelet:

[0030]

[0031] Among them, σ 2 Represents the width parameter of the mother wavelet.

[0032] Based on the further improvement of the above working state detection method, the pre-target improved AlexNet model includes a first input layer, a first convolutional pooling layer, a second convolutional pooling layer and a third convolutional pooling layer, as well as a second input layer, a fusion full-link layer, a first full-link layer, a second full-link layer and a third full-link layer;

[0033] The target working feature image passes through the first input layer, the first convolutional pooling layer, the second convolutional pooling layer and the third convolutional pooling layer in sequence to obtain a first target working feature value;

[0034] The target working characteristic vector passes through the second input layer, and is expanded by a preset multiple in the second input layer to obtain a second target working characteristic value;

[0035] The fusion full-link layer is used to concatenate and fuse the first target working characteristic value and the second target working characteristic value to obtain a concatenated and fused target working characteristic value;

[0036] The spliced ​​and fused target working feature values ​​are sequentially passed through the first full-link layer, the second full-link layer and the third full-link layer to obtain the second target working state prediction result when the target device is working.

[0037] Based on further improvement of the above working status detection method, the second preset multiple is 2-4.

[0038] Based on the further improvement of the above working state detection method, the fusion decision neural network is constructed according to the Hecht-Nelson method, and the fusion decision neural network includes two input layers, a preset number of hidden layers and an output layer;

[0039] The two input layers are used to receive the first target working state prediction result and the second target working state prediction result respectively, and each input layer includes M neurons; wherein the types of working states of the target device are also M;

[0040] Each hidden layer consists of 2M+1 neurons, and the output layer consists of M neurons.

[0041] Based on the further improvement of the above working status detection method, the preset number of layers is 2-3;

[0042] The obtaining of the target working state detection result also includes:

[0043] The fused target working state prediction result is obtained from the fused decision neural network, and the prediction category with the highest probability is screened out from the fused target working state prediction result as the final target working state detection result.

[0044] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0045] 1. By extracting the preset features of the original time domain amplitude signal and the first-order differential signal of the time domain amplitude, a target working state prediction result is obtained using the target support vector classifier model; at the same time, the target working feature image obtained by continuous wavelet transform of the original time domain amplitude signal is used, and another working state detection result is obtained using the target improved AlexNet model; finally, the two target working state prediction results are fused using the fusion decision neural network to obtain the final target working state detection result, which uses the electromagnetic signals generated by the target device when it is working in multiple dimensions, making the final prediction result more accurate, so as to accurately monitor the working state of the target device in real time;

[0046] 2. Using the improved kernel function in the target support vector classifier model can accelerate the training of the target support vector classifier model and make the trained target support vector classifier model predict the target working feature vector more accurately;

[0047] 3. The original time domain amplitude signal is transformed into a continuous wavelet using the improved mother wavelet, so that the target working characteristic image is richer; and the target is improved by using the Alexnet model, which can use the one-dimensional target working characteristic vector and the two-dimensional target working characteristic image for fusion prediction, thereby improving the prediction accuracy of the working status of the target device;

[0048] 4. The prediction results of the target working feature vector and the prediction results of the target working feature vector and the target working feature image are fused by using the fusion decision neural network, and the various characteristics of the original time domain amplitude signal are comprehensively measured to obtain the final target working state detection result with higher reliability.

[0049] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0051] Figure 1 A schematic diagram of a process for detecting the working state of a target device provided by an embodiment of the present invention;

[0052] Figure 2 A schematic diagram of the structure of the target improved AlexNet model provided by an embodiment of the present invention;

[0053] Figure 3 A schematic diagram of the structure of a fusion decision neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0055] A specific embodiment of the present invention discloses a method for detecting the working status of a target device, such as Figure 1 As shown, the working status detection method includes:

[0056] Step S1: obtaining an original time domain amplitude signal and a first-order differential signal of the time domain amplitude when the target device is working, extracting preset features of the original time domain amplitude signal and the first-order differential signal of the time domain amplitude respectively, and obtaining a target working feature vector when the target device is working; performing a continuous wavelet transform on the original time domain amplitude signal to obtain a target working feature image;

[0057] Step S2: inputting the target working feature vector into the trained target support vector classifier model to obtain a first target working state prediction result when the target device is working;

[0058] Step S3: inputting the target working feature vector and the target working feature image into the pre-trained target improved AlexNet model to obtain a second target working state prediction result when the target device is working;

[0059] Step S4: Based on the pre-constructed fusion decision neural network, the first target working state prediction result and the second target working state prediction result are fused to obtain the final target working state detection result.

[0060] Specifically, the target device refers to an electronic device that has the ability to radiate and emit electromagnetic signals when working, such as detecting the circuit of a closed system or detecting a radar system. The target device can include multiple working states when working, and for different target devices, the working states need to be divided according to actual conditions.

[0061] Specifically, Figure 1 As shown, in order to detect the working state of the target device, in step S1, the original time domain amplitude signal of the target device when it is working is collected. The original time domain amplitude signal is an electromagnetic signal emitted by the target device when it is working, and the amplitude of the electromagnetic signal changes with time.

[0062] Specifically, the time domain amplitude first-order differential signal is a signal obtained by first-order derivative based on the original time domain amplitude signal. After obtaining the original time domain amplitude signal and the time domain amplitude first-order differential signal, the included preset features are extracted respectively, and the extracted preset features are combined as the target working feature vector corresponding to the current working of the target device.

[0063] Preferably, the preset features include one or more of the following:

[0064] Maximum value;

[0065] Minimum value;

[0066] Crest factor;

[0067] Zero crossing rate;

[0068] Mean square error.

[0069] Specifically, the features extracted from the original time-domain amplitude signal and the features extracted from the time-domain amplitude first-order differential signal may be the same or different. Preferably, the same preset features are extracted from the original time-domain amplitude signal and the time-domain amplitude first-order differential signal. For example, the features of maximum value, minimum value, peak factor, zero-crossing rate and mean square error are extracted from the original time-domain amplitude signal to obtain 5 features corresponding to the original time-domain amplitude signal; at the same time, the features of maximum value, minimum value, peak factor, zero-crossing rate and mean square error are extracted from the time-domain amplitude first-order differential signal to obtain 5 features corresponding to the time-domain amplitude first-order differential signal. After combination, the target working feature vector obtained when the target device is working includes 10 features.

[0070] Specifically, Figure 1 As shown, in step S1, a continuous wavelet transform is performed on the original time domain amplitude signal to obtain a target working characteristic image.

[0071] Preferably, the original time domain amplitude signal is subjected to continuous wavelet transform by the following formula to obtain the target working characteristic image:

[0072]

[0073] Among them, s(t) represents the original time domain amplitude signal, S(a,b) represents the target working characteristic image, a represents the translation position, b represents the scaling factor, ψ(t) represents the mother wavelet, represents complex conjugate.

[0074] Specifically, a continuous wavelet transform is performed on the original time domain amplitude signal when the target device is working to obtain a two-dimensional image. The continuous wavelet transform can construct a time-frequency signal with good time domain and frequency domain data parts.

[0075] Specifically, the interval of a is (-∞, +∞), and the interval of b is (0, +∞); from the perspective of time-frequency analysis, the larger the b value, the smaller the change in frequency; from the perspective of time-frequency analysis, the smaller the b value, the larger the change in frequency.

[0076] Specifically, the mother wavelet may be a Morlet wavelet, a Haar wavelet or a Daubechies wavelet.

[0077] Preferably, the following function is used as the mother wavelet:

[0078]

[0079] Among them, σ 2 Represents the width parameter of the mother wavelet.

[0080] Specifically, σ 2 It is a custom width parameter, which can be set reasonably according to the actual situation.

[0081] Specifically, Figure 1 As shown, in step S2, the target working feature vector is input into the trained target support vector classifier model to obtain a first target working state prediction result when the target device is working.

[0082] It is worth noting that the target support vector classifier model needs to be trained in advance; specifically, it includes:

[0083] Construct training and test sets:

[0084] Determine the target device, set the type of working state of the target device, so that the target device works in different types of working states respectively, collect the same number of original time domain amplitude signals in different working states, and perform first-order differential derivative on each original time domain amplitude signal to obtain the corresponding time domain amplitude first-order differential signal; extract the preset features of each original time domain amplitude signal and the corresponding time domain amplitude first-order differential signal respectively as the target working feature vector corresponding to each original time domain amplitude signal; take the target working feature vector of each original time domain amplitude signal and the type of corresponding working state as a sample, and divide all samples into a training set and a test set according to a certain ratio;

[0085] Training of target support vector classifier model:

[0086] The target support vector classifier model is trained using samples of the training set, and the trained target support vector classifier model is then tested using the test set to obtain the final target support vector classifier model.

[0087] The target support vector classifier model that is finally trained can obtain the corresponding working state according to the electromagnetic signal emitted by the target device when it is working.

[0088] Preferably, the kernel function used in the training of the target support vector classifier model is:

[0089]

[0090] Among them, x i 、x j represents any two training samples, σ 1 Represents the hyperparameters of the target support vector classifier model.

[0091] Specifically, the kernel function satisfies the symmetry and Mercer conditions and has only one hyperparameter σ 1 , σ 1 It is a positive real constant with a simple function form, which is suitable for the state classification of the target device when it is working.

[0092] Specifically, choose appropriate penalty parameter C and hyperparameter σ 1 , a larger C value will lead to a better fit to the training data, but may overfit; a smaller C value will lead to stronger regularization.

[0093] Preferably, the optimization conditions satisfied during the training of the target support vector classifier model are:

[0094]

[0095] Among them, N is the total number of training set samples, α iis the Lagrange multiplier of the i-th sample, y i is the category label value of the i-th sample, and C is the penalty parameter;

[0096] Under the premise of the optimization conditions, the following objective function is solved to obtain the Lagrange multiplier of each sample corresponding to the minimum value of the objective function, and the target support vector classifier model training is completed:

[0097]

[0098] Among them, α j is the Lagrange multiplier of the jth sample, y j is the class label value of the jth sample.

[0099] Specifically, the training process of the target support vector classifier model can be connected to the process of solving the dual problem. The support vector machine classifier is a convex optimization problem, and the solution process is to find the Lagrange multiplier. However, due to its constraint condition being the number of training samples, the general optimization algorithm is relatively expensive, and the sequential minimum optimization algorithm is preferably used.

[0100] Specifically, Figure 1 As shown, in step S2, according to the target working feature vector of the target device when working, combined with the target support vector classifier model, the prediction result of the target device when working is obtained. The target device includes M working states when working, and the first target working state prediction result includes the probability corresponding to each working state, that is, S = [S 1 ,S 2 ,...S m ...,S M ],in, 0≤S m ≤1,m∈{1,2,...,M}, S represents the prediction result of the first target working state, S m Represents the probability of the mth working state.

[0101] Specifically, Figure 1 As shown, in step S3, the target working feature vector and the target working feature image obtained in step S1 are input into the pre-trained target improved AlexNet model to obtain a second target working state prediction result when the target device is working.

[0102] It is understandable that the target improved AlexNet model also needs to be trained before use. The process of obtaining the training set refers to the training process of the target support vector classifier. The difference is that in addition to the target working feature vector, the training sample set of the target improved AlexNet model also needs to perform continuous wavelet transform on the original time domain amplitude signal to obtain the target working feature image; during training, the target working feature vector and the target working feature image are used as input data, and the corresponding working state is used as a label to train the model.

[0103] Preferably, if Figure 2 As shown, the target improved AlexNet model includes a first input layer, a first convolutional pooling layer, a second convolutional pooling layer, and a third convolutional pooling layer, as well as a second input layer, a fusion full-link layer, a first full-link layer, a second full-link layer, and a third full-link layer;

[0104] The target working feature image passes through the first input layer, the first convolutional pooling layer, the second convolutional pooling layer and the third convolutional pooling layer in sequence to obtain a first target working feature value;

[0105] The target working characteristic vector passes through the second input layer, and is expanded by a preset multiple in the second input layer to obtain a second target working characteristic value;

[0106] The fusion full-link layer is used to concatenate and fuse the first target working characteristic value and the second target working characteristic value to obtain a concatenated and fused target working characteristic value;

[0107] The spliced ​​and fused target working feature values ​​are sequentially passed through the first full-link layer, the second full-link layer and the third full-link layer to obtain the second target working state prediction result when the target device is working.

[0108] Specifically, Figure 2 As shown, the target improved AlexNet model includes the first input layer, the first convolutional pooling layer, the second convolutional pooling layer and the third convolutional pooling layer from top to bottom, wherein the first convolutional pooling layer includes the first convolutional layer and the first maximum pooling layer, the second convolutional pooling layer includes the second convolutional layer and the second maximum pooling layer, and the third convolutional pooling layer includes the third convolutional layer, the fourth convolutional layer, the fifth convolutional layer and the third maximum pooling layer. The target working feature image is input from the first input layer, and the first target working feature value is obtained by passing through the first convolutional pooling layer, the second convolutional pooling layer and the third convolutional pooling layer in sequence.

[0109] Specifically, Figure 2As shown, the target improved AlexNet model also includes a second input layer, a fusion full link layer, a first full link layer, a second full link layer, and a third full link layer. The target working feature vector is input into the target improved AlexNet model through the second input layer, and is expanded by a preset multiple in the second input layer to obtain a second target working feature value.

[0110] Preferably, the second preset multiple is 2-4. It is worth noting that the target working feature vector input through the second input layer is extended in sequence to obtain a feature vector of the second preset multiple. Exemplarily, the target working feature vector input through the second input layer is copied to obtain a feature vector of the second preset multiple, and the feature vectors of the second preset multiple are spliced ​​to obtain the second target working feature value.

[0111] Specifically, Figure 2 As shown, in the fusion full link layer, the first target working characteristic value and the second target working characteristic value are spliced ​​and fused to obtain the spliced ​​and fused target working characteristic value. The spliced ​​and fused target working characteristic value passes through the first full link layer, the second full link layer and the third full link layer in turn to obtain the second target working state prediction result when the target device is working, that is, S'=[S' 1 ,S' 2 ,...S' m ...,S' M ],in, 0≤S' m ≤1,m∈{1,2,...,M}, S' represents the prediction result of the second target working state, S' m Represents the probability of the mth working state.

[0112] Specifically, Figure 2 As shown in the figure, the first input layer is the image input layer, the first convolution layer and the first maximum pooling layer are convolution layers and maximum pooling, which are used to extract low-level features in the input signal image, such as edges and textures. 96 11x11 convolution kernels with a stride of 4 can be used to output a 55x55x96 feature map, followed by ReLU activation and maximum pooling, using a 3x3 convolution kernel with a stride of 2.

[0113] The second convolution layer and the second maximum pooling layer are convolution layers and maximum pooling, which are used to further extract complex features in the signal image. 256 5x5 convolution kernels can be used with a stride of 1 to output a 27x27x256 feature map, followed by ReLU activation and maximum pooling, using a 3x3 convolution kernel with a stride of 2.

[0114] The third convolution layer, the fourth convolution layer and the fifth convolution layer are all convolution layers, which are used for higher features in the signal image. 384 3x3 convolution kernels can be used with a stride of 1 to output a 13x13x384 feature map. The third maximum pooling layer is the maximum pooling, which is used for feature downsampling and feature selection. The maximum pooling uses a 3x3 convolution kernel with a stride of 2.

[0115] The fusion fully-linked layer is a fusion feature fully-linked layer, which is used to fuse the first target working feature value and the second target working feature value of the signal, expand the features extracted by the convolutional layer, perform feature combination and nonlinear transformation, output 4096 neurons, and perform ReLU activation.

[0116] The first and second fully connected layers further perform feature combination and nonlinear transformation, input a vector, output 4096 neurons, and perform ReLU activation.

[0117] The third fully connected layer is a fully connected layer, which is used to output the classification results. It inputs a vector, outputs M neurons, and performs softmax activation to obtain the probability distribution of M categories, that is, the prediction result of the second target working state when the target device is working.

[0118] Specifically, Figure 1 As shown, in step S4, the first target working state prediction result obtained in step S2 and the second target working state prediction result obtained in step S3 are input into a pre-constructed fusion decision neural network, and the first target working state prediction result and the second target working state prediction result are fused in the pre-constructed fusion decision neural network to obtain the final target working state detection result.

[0119] Preferably, the fusion decision neural network is constructed according to the Hecht-Nelson method, and the fusion decision neural network includes two input layers, a preset number of hidden layers and an output layer;

[0120] The two input layers are used to receive the first target working state prediction result and the second target working state prediction result respectively, and each input layer includes M neurons; wherein the types of working states of the target device are also M;

[0121] Each hidden layer consists of 2M+1 neurons, and the output layer consists of M neurons.

[0122] Specifically, Figure 3 As shown, the first target working state prediction result and the second target working state prediction result are input into the two input layers at the same time, with a total input of 2M; each hidden layer includes 2M+1 neurons, and the output layer includes M neurons. It can be understood that M is the type of working state of the target device.

[0123] Preferably, the preset number of layers is 2-3;

[0124] The obtaining of the target working state detection result also includes:

[0125] The fused target working state prediction result is obtained from the fused decision neural network, and the prediction category with the highest probability is screened out from the fused target working state prediction result as the final target working state detection result.

[0126] Specifically, Figure 3 As shown, the fusion decision neural network includes 2 hidden layers.

[0127] Specifically, the target working state prediction result obtained by fusion of the first target working state prediction result and the second target working state prediction result is obtained from the fusion decision neural network. The fused target working state prediction result is S”=[S” 1 ,S” 2 ,...S” m ...,S” M ],in, 0≤S” m ≤1,m∈{1,2,...,M}, S” represents the prediction result of the target working state after fusion, S” m Represents the probability of the mth working state.

[0128] Compare S" 1 , S” 2 ,...S” m ...、S” M , find the predicted category with the highest probability as the working state corresponding to the current work of the target device.

[0129] Compared with the prior art, a method for detecting the working state of a target device provided by an embodiment of the present invention obtains a target working state prediction result by extracting preset features of an original time-domain amplitude signal and a first-order differential signal of the time-domain amplitude, and using a target support vector classifier model; at the same time, a target working feature image is obtained by performing continuous wavelet transform on the original time-domain amplitude signal, and another working state detection result is obtained by using a target improved AlexNet model; finally, a fusion decision neural network is used to fuse the two target working state prediction results to obtain a final target working state detection result, and the electromagnetic signals generated by the target device when it is working are used in multiple dimensions, so that the final prediction result is more accurate, so as to facilitate real-time and precise monitoring of the working state of the target device; in the target support vector classifier model, the target working feature image is obtained by performing continuous wavelet transform on the original time-domain amplitude signal, and another working state detection result is obtained by using a target improved AlexNet model; finally, the two target working state prediction results are fused by using a fusion decision neural network, and a final target working state detection result is obtained, and the electromagnetic signals generated by the target device when it is working are used in multiple dimensions, so that the final prediction result is more accurate, so as to facilitate real-time and precise monitoring of the working state of the target device; in the target support vector classifier model, the target working state prediction result is obtained by using a target working state prediction result. The improved kernel function can accelerate the training of the target support vector classifier model, and make the trained target support vector classifier model more accurate in predicting the target working feature vector; the improved mother wavelet is used to perform continuous wavelet transform on the original time domain amplitude signal, so that the obtained target working feature image is richer; and the target improved Alexnet model is used to predict the target working feature vector and the target working feature image, thereby improving the prediction accuracy of the target device working state; the fusion decision neural network is used to fuse the prediction results of the target working feature vector, as well as the prediction results of the target working feature vector and the target working feature image, thereby comprehensively measuring various aspects of the original time domain amplitude signal to obtain the final target working state detection result with higher reliability.

[0130] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0131] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for detecting the working status of a target device, characterized in that: The working status detection method comprises: Acquire the original time domain amplitude signal and the first-order differential signal of the time domain amplitude when the target device is working, extract the preset features of the original time domain amplitude signal and the first-order differential signal of the time domain amplitude respectively, and obtain the target working feature vector when the target device is working; perform continuous wavelet transform on the original time domain amplitude signal to obtain the target working feature image; Inputting the target working feature vector into the trained target support vector classifier model to obtain a first target working state prediction result when the target device is working; Input the target working feature vector and the target working feature image into the pre-trained target improved AlexNet model to obtain the second target working state prediction result when the target device is working; Based on the pre-built fusion decision neural network, the first target working state prediction result and the second target working state prediction result are fused to obtain the final target working state detection result.

2. The working status detection method according to claim 1, characterized in that: Preset characteristics include one or more of the following: Maximum value; Minimum value; Crest factor; Zero crossing rate; Mean square error.

3. The working status detection method according to claim 2, characterized in that: The kernel function used in the training of the target support vector classifier model is: Among them, x i 、x j represents any two training samples, and σ1 represents the hyperparameter of the target support vector classifier model.

4. The working status detection method according to claim 3, characterized in that: The optimization conditions satisfied during the training of the target support vector classifier model are: Among them, N is the total number of training set samples, α i is the Lagrange multiplier of the i-th sample, y i is the category label value of the i-th sample, and C is the penalty parameter; Under the premise of the optimization conditions, the following objective function is solved to obtain the Lagrange multiplier of each sample corresponding to the minimum value of the objective function, and the target support vector classifier model training is completed: Among them, α j is the Lagrange multiplier of the jth sample, y j is the class label value of the jth sample.

5. The working status detection method according to claim 1, characterized in that: The original time domain amplitude signal is transformed by continuous wavelet transform using the following formula to obtain the target working characteristic image: Among them, s(t) represents the original time domain amplitude signal, S(a,b) represents the target working characteristic image, a represents the translation position, b represents the scaling factor, ψ(t) represents the mother wavelet, represents complex conjugate.

6. The working status detection method according to claim 5, characterized in that: The following function is used as the mother wavelet: Among them, σ2 represents the width parameter of the mother wavelet.

7. The working status detection method according to claim 1, characterized in that: The target improved AlexNet model includes the first input layer, the first convolutional pooling layer, the second convolutional pooling layer and the third convolutional pooling layer, as well as the second input layer, the fusion full connection layer, the first full connection layer, the second full connection layer and the third full connection layer; The target working feature image passes through the first input layer, the first convolutional pooling layer, the second convolutional pooling layer and the third convolutional pooling layer in sequence to obtain a first target working feature value; The target working characteristic vector passes through the second input layer, and is expanded by a preset multiple in the second input layer to obtain a second target working characteristic value; The fusion full-link layer is used to concatenate and fuse the first target working characteristic value and the second target working characteristic value to obtain a concatenated and fused target working characteristic value; The spliced ​​and fused target working feature values ​​are sequentially passed through the first full-link layer, the second full-link layer and the third full-link layer to obtain the second target working state prediction result when the target device is working.

8. The working status detection method according to claim 7, characterized in that: The second preset multiple is 2-4.

9. The working status detection method according to claim 1, characterized in that: The fusion decision neural network is constructed according to the Hecht-Nelson method, and the fusion decision neural network includes two input layers, a preset number of hidden layers and an output layer; The two input layers are used to receive the first target working state prediction result and the second target working state prediction result respectively, and each input layer includes M neurons; wherein the types of working states of the target device are also M; Each hidden layer consists of 2M+1 neurons, and the output layer consists of M neurons.

10. The working status detection method according to claim 9, characterized in that: The preset number of layers is 2-3; The obtaining of the target working state detection result also includes: The fused target working state prediction result is obtained from the fused decision neural network, and the prediction category with the highest probability is screened out from the fused target working state prediction result as the final target working state detection result.

Citation Information

Patent Citations

  • Brushless direct current motor sensor fault detection method based on convolutional neural network

    CN112036435A

  • Power system fault diagnosis information fusion method and device based on deep neural network

    CN112285484A

  • Photovoltaic inverter fault diagnosis method based on wavelet time-frequency diagram

    CN114021598A

  • Semiconductor process prediction method and device considering overall features and local features

    CN114819242A

  • Differential data enhancement-based information interaction fusion gearbox fault diagnosis method

    CN115165347A