Fault diagnosis method, system, medium and device based on part feature frequency

Abstract

The present application relates to the field of mechanical equipment fault diagnosis, and discloses a fault diagnosis method, system, medium and equipment based on part characteristic frequency, which comprises the following steps: obtaining part fault characteristic frequency from a mechanical equipment original fault data set, and constructing guidance information of physical knowledge based on the part fault characteristic frequency; splicing the generated fault characteristic frequency information and fault data to form double-channel data, inputting the double-channel data into a multi-output deep convolutional neural network for training to extract fault features; performing multi-output classification and diagnosis on the fault features by the multi-output deep convolutional neural network, each output corresponding to a fault type of a mechanical equipment, extracting fault features corresponding to the fault type from global features to complete diagnosis on whether the fault exists in the mechanical equipment. The present application can decouple composite faults and realize prediction of single fault types.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical equipment fault diagnosis, and in particular to a fault diagnosis method, system, medium and equipment based on part characteristic frequency. BACKGROUND

[0002] Traditional fault diagnosis technology is usually based on theoretical analysis, combined with the operating principle and physical characteristics of the equipment, to extract characteristic parameters associated with faults. Taking rotating machinery as an example, the frequency characteristics of the vibration signal are key analysis indicators. When a bearing fails, specific fault characteristic frequency components will appear in its vibration spectrum, and by identifying these frequency components, the health status of the equipment can be effectively determined. However, in actual engineering applications, due to factors such as manufacturing errors, mechanical wear during operation, and speed fluctuations, the characteristic frequencies in the measured signal may exhibit slight shifts. Coupled with background noise and some equipment-specific vibration patterns, experienced experts are needed for detailed analysis and determination.

[0003] However, in the field of deep neural network fault diagnosis, when faced with complex and variable fault signals, it is difficult to accurately predict single-class faults. Moreover, due to the lack of practical physical knowledge, the model cannot correctly extract fault features on a small fault data set. SUMMARY

[0004] To solve the above problems, the present application aims to provide a fault diagnosis method, system, medium and equipment based on part characteristic frequency, which can decouple complex faults and predict single fault types.

[0005] To achieve the above-mentioned purpose, in the first aspect, the technical solution adopted by the present application is as follows: a fault diagnosis method based on part characteristic frequency, comprising: obtaining part fault characteristic frequencies from the original fault data set of the mechanical equipment, and constructing guidance information of physical knowledge based on the part fault characteristic frequencies; concatenating the generated fault characteristic frequency information with the fault data to form dual-channel data, and inputting the dual-channel data into a multi-output deep convolutional neural network for training to extract fault features; performing multi-output classification and diagnosis of the fault features by the multi-output deep convolutional neural network, each output corresponding to a fault type of the mechanical equipment, extracting the fault features corresponding to the fault type from the global features, and completing the diagnosis of whether the fault exists in the mechanical equipment.

[0006] Further, the part fault characteristic frequencies are obtained from the original fault data set of the mechanical equipment, and the guidance information of physical knowledge is constructed based on the part fault characteristic frequencies, which includes:

[0007] Based on the failure principle of the part and the parameter information of the part, a part failure characteristic frequency is calculated, and a characteristic frequency band is formed by spreading to the left and right of the part failure characteristic frequency as the center;

[0008] The collected vibration data of the mechanical equipment is band-pass filtered by using the characteristic frequency band, and the obtained failure characteristic frequency is guided information of physical knowledge.

[0009] Further, the generated failure characteristic frequency information is spliced with the failure data to form double-channel data, including:

[0010] According to the different parts and needs, the guided information is spliced with the originally collected vibration data, and then frequency spectrum conversion is performed to convert to the frequency domain or time-frequency domain to form double-channel data.

[0011] Further, the multi-output deep convolutional neural network adopts a parallel convolutional network, and the number of parallel convolutional networks is determined according to the part data or the type of the frequency spectrum signal.

[0012] Further, the multi-output deep convolutional neural network is used for multi-output classification of the failure characteristics, including:

[0013] The extracted multi-dimensional complex failure characteristics are converted into one-dimensional features by using a flattening layer, or dimension reduction is performed by using a global average pooling layer;

[0014] A plurality of output branches are connected, and each branch corresponds to a failure type; in each branch, N layers of fully connected layers are selected according to needs, and finally a fully connected layer with only one output and using a sigmoid activation function is used to output the probability of the failure, to determine whether the failure exists.

[0015] Further, after the multi-output classification of the failure, a step of constraint is further included:

[0016] All single-failure probabilities of the outputs are spliced together, and multi-output constraint is performed to further constrain each independent branch output.

[0017] Further, the multi-output constraint is that the F1 score and the cross-entropy loss in the convolutional network are used to further constrain each independent branch output.

[0018] In a second aspect, the technical scheme adopted by the present application is: a failure diagnosis system based on part characteristic frequency, comprising:

[0019] A physical knowledge information acquisition module acquires part failure characteristic frequencies from a mechanical equipment original failure data set, and constructs guided information of physical knowledge based on the part failure characteristic frequencies;

[0020] The feature extraction module splices the generated fault feature frequency information with fault data to form double-channel data, and inputs the double-channel data into a deep convolutional neural network with multiple outputs for training, so as to extract fault features.

[0021] The fault decoupling classification and diagnosis module performs multiple output classification and diagnosis on the fault features by using the deep convolutional neural network with multiple outputs, each output corresponds to a fault type of the mechanical equipment, the fault features corresponding to the fault type are extracted from the global features, and whether the fault exists in the mechanical equipment is diagnosed.

[0022] In a third aspect, a computer-readable storage medium storing one or more programs includes instructions that, when executed by a computing device, cause the computing device to perform any of the methods described above.

[0023] In a fourth aspect, a computing device includes one or more processors, memory, and one or more programs stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described above.

[0024] The present application has the following advantages due to the above technical solutions:

[0025] 1. The present application extracts part fault feature frequency from the original fault data set of the mechanical equipment, and constructs guidance information of physical knowledge based on the part fault feature frequency, effectively solving the problem that the model cannot correctly extract fault features on a small fault data set due to the lack of actual physical knowledge.

[0026] 2. The present application performs multiple output classification and diagnosis on the composite fault features by using the deep convolutional neural network with multiple outputs, and combines multiple output constraints to decouple the composite fault, even if the corresponding composite fault data is not seen in the training stage, the running state of the mechanical equipment is accurately diagnosed, when there are multiple faults in the system, the multiple faults can be decomposed and diagnosed as multiple individual faults, and overfitting to single fault features is effectively avoided. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is the overall flowchart of the fault diagnosis method based on part feature frequency in the embodiment of the present application;

[0028] Figure 2 is the detailed flowchart of the fault diagnosis method based on part feature frequency in the embodiment of the present application;

[0029] Figure 3is the envelope spectrum of the vibration signal in the compound fault feature of the bearing outer ring fault and the flywheel imbalance in the embodiment of the application;

[0030] Figure 4 is the feature frequency information generated based on the envelope spectrum in the compound fault feature of the bearing outer ring fault and the flywheel imbalance in the embodiment of the application;

[0031] Figure 5 is the compound fault classification and diagnosis flowchart of the bearing in the embodiment of the application;

[0032] Figure 6 is the deep neural network fault diagnosis schematic diagram based on a customized wavelet convolution kernel in the prior art scheme one;

[0033] Figure 7 is the fault diagnosis schematic diagram of the deep learning compound fault diagnosis model in the prior art scheme two;

[0034] Figure 8 is the fault diagnosis schematic diagram of the deep learning compound fault diagnosis model in the prior art scheme three. DETAILED DESCRIPTION

[0035] Due to the end-to-end adaptive learning and feature extraction, deep neural networks have been widely and deeply studied in the field of mechanical part fault diagnosis. The current development of deep learning technology highly depends on the support of large-scale labeled data, especially in the fields of image classification and natural language processing, a mature data labeling system has been formed. However, in the field of mechanical fault diagnosis, due to the complex equipment operating environment, high cost and long cycle of fault sample collection, the amount of available labeled data is extremely limited. This data scarcity makes the model prone to overfitting during training, and the model tends to capture non-essential features (such as noise interference or accidental states of equipment operation) in the training samples. Although high accuracy can be achieved on a specific test set, the generalization ability in cross-condition and cross-device scenarios is significantly reduced. This surface feature-based learning mode not only limits the practical application value of the diagnosis model, but also raises doubts about the reliability of its diagnosis results.

[0036] In the actual operation of the device, composite failure has become a common failure, which has high uncertainty and low predictability. Device failure not only leads to a decrease in economic benefits, but also can affect personnel and the environment. Therefore, studying general and efficient mechanical part multi-fault diagnosis technology is of great significance to ensure the safety of the production process and prevent and avoid accidents. In the diagnosis and processing of composite faults, many scholars have conducted relevant research and exploration. Early studies used softmax activation function as the output of the convolutional network model, treating composite failure as a new fault mode rather than a coupling mode of single failure. Using softmax as a classifier can only treat composite failure as a new fault mode, but cannot decouple single failure into composite failure, which leads to a sharp increase in classification labels, and because the model training is difficult to cover all composite failure combinations, it cannot be applied in practice.

[0037] With further research, many studies have started to use multi-label classifiers to decouple composite failure due to the shortcomings of softmax. Unlike multi-class classifiers using softmax, multi-label classifiers use sigmoid activation function as the activation function of the last layer of the classifier. Using the sigmoid activation function, each output probability of the model is independent of each other, so the fault classification label can be output according to the number of single faults in the composite failure, realizing the decoupling of the composite failure. Compared with softmax, the multi-label classifier using sigmoid can complete the decoupling of the composite failure and will not cause an increase in the classification label. However, due to the limitations of the network structure, the network model using the sigmoid multi-label classifier cannot predict the composite failure combinations that have not been seen in the training phase, even if all the features of the single failure have been learned in the training phase. Therefore, when training the model using the sigmoid classifier, the model needs to be provided with all the composite failure data expected to be predicted. Therefore, in existing research, there are mainly the following schemes:

[0038] Scheme one: integrating physical knowledge into deep neural networks is currently a mainstream direction, that is, using physical knowledge to guide the learning of neural network models to avoid learning non-essential features. The main idea of improving the CNN network architecture to enhance the feature learning ability of the fault diagnosis model is to replace the traditional convolution kernel with a kernel with physical meaning. Typical methods such as custom wavelet convolution kernel, as shown in Figure 6 For example, using a continuous wavelet convolution layer instead of the first convolution layer of the CNN model is equivalent to adding some constraints to the waveforms of the convolution kernel. Therefore, it can extract feature information with clear physical meaning from the data. This mechanism can enable the CNN model to achieve efficient feature selection and reveal the correlation mechanism between the random feature space and the signal space.

[0039] But an important aspect of the scheme is the selection of wavelet basis. Different wavelet bases have different properties, and the selection of wavelet basis directly affects the decomposition effect of the signal, the denoising effect, etc. Therefore, in practical application, when facing complex and variable fault signals, the adaptability and universality of the wavelet convolution kernel need to be improved.

[0040] As shown in the second scheme, a softmax activation function is used in the last fully connected layer of the fault diagnosis classification module. The softmax activation function is usually used in the multi-classification scenario, that is, there are multiple classification types in the scenario, and a sample belongs to only one classification, and different classifications are mutually exclusive. The softmax maps the outputs of multiple neurons to the [0, 1] interval according to the following formula, and the sum of all outputs is 1. Figure 7

[0041] In the formula, z i represents the output value of the i-th node; j is the number of output nodes, that is, the number of classification categories.

[0042] In this multi-classification scenario, the softmax function predicts the probability of each category, and usually selects the category with the highest predicted value as the result. That is, each prediction only outputs one predicted category as the prediction result. In this scenario, softmax cannot output multiple predicted categories at the same time. Therefore, when the composite fault classifier uses this scheme, it regards a composite fault as a label, that is, as a single fault category.

[0043] It can be seen that the scheme has the following problems: 1. The biggest problem is that the composite fault is regarded as a single fault. There are many parts in mechanical equipment, and there are many fault types. At this time, the number of composite fault combinations is huge, and different application scenarios and working conditions cannot be applied in practice. 2. Because the number of composite fault combinations is large, the data collection for training the model faces great challenges, and it is impossible to realize in practice.

[0044] As shown in the third scheme, a sigmoid activation function is used in the last fully connected layer of the fault diagnosis classification module. The sigmoid activation function is usually used in the multi-label classification scenario, that is, a sample can belong to multiple label classification types, and the categories are not mutually exclusive, such as different types of cars, sedans, SUVs, trucks, etc. The sigmoid maps the outputs of multiple neurons to the [0, 1] interval according to the following formula.

[0045] Figure 8

[0046]

[0047] ​​In the formula, x represents an output value of a neuron.

[0048] In this multi-label classification scenario, the sigmoid function predicts the probability of each class, and when the probability is greater than a certain threshold, such as 0.5, it is considered that the sample belongs to this class. At this time, the model can output multiple classification labels as the prediction result each time. In this scenario, the sigmoid can output multiple classes at the same time. Therefore, when the composite fault classifier adopts this scheme, it can theoretically predict a composite fault as multiple labels, each of which corresponds to a single fault class.

[0049] However, the biggest problem of this scheme is that in actual application, to identify a composite fault as multiple single fault classifications, the model must have seen the data of this composite fault during training and complete the learning process according to the label. If the model has not seen the data of a composite fault during training, it cannot correctly make a classification diagnosis in actual application. Moreover, due to the numerous combinations of composite faults, the collection of data for training the model faces great challenges, and it is impossible to realize in practice.

[0050] Therefore, in order to solve the problems in the prior art and the above schemes, the present application provides a fault diagnosis method, system, medium and equipment based on part feature frequency, which introduces prior knowledge of mechanical part fault feature frequency, constructs a theoretically guided feature constraint mechanism, guides the neural network to focus on the feature frequency components directly related to the fault mode during the training process, and thereby suppresses the interference of irrelevant features. At the same time, a multi-output classification scheme in the convolutional neural network is adopted, each output corresponds to a single fault type of a mechanical equipment, whether the specified single fault feature exists is searched according to all the features extracted by the neural network, and the result whether the corresponding single fault type exists is output.

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present application.

[0052] It should be noted that the terms used herein are only intended to describe specific embodiments, and are not intended to limit the exemplary embodiments according to the present application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the specification, there is a feature, step, operation, device, component, and / or combination thereof.

[0053] In one embodiment of the present application, a fault diagnosis method based on part characteristic frequency is provided. In this embodiment, as shown in Figure 1 、 Figure 2 , the method comprises the following steps:

[0054] 1) Extracting part fault characteristic frequency from the original fault data set of the mechanical equipment, constructing guidance information of physical knowledge based on the part fault characteristic frequency, for training the neural network model, and guiding the neural network to focus on the characteristic frequency components directly related to the fault mode in the training process;

[0055] Wherein, the guidance information of physical knowledge refers to the generated fault characteristic frequency. The generation process of this characteristic frequency is as follows: the part fault characteristic frequency is calculated based on the fault principle of the part and the parameter information of the part, such as the pitch diameter of a certain type of bearing and the diameter of the rolling body, etc. Then, based on this calculated characteristic frequency, a certain bandwidth, such as 5Hz, is formed into a characteristic frequency band. The characteristic frequency band is used to filter the vibration information collected from the equipment to obtain the guidance information. Therefore, it is called guidance information based on physical knowledge, which can also be called guidance information of fault characteristic frequency.

[0056] 2) Splicing the generated fault characteristic frequency information with the fault data to form double-channel data, inputting the multi-output deep convolutional neural network for training to extract fault features; wherein, the fault features include composite fault features and single fault features;

[0057] 3) Multi-output classification and diagnosis of the fault features by the multi-output deep convolutional neural network, each output corresponding to a fault type of the mechanical equipment, extracting the fault features corresponding to the fault type from the global features to complete the diagnosis of whether the fault exists in the mechanical equipment.

[0058] In the above step 1), the part fault characteristic frequency is obtained from the original fault data set of the mechanical equipment, and the guidance information of physical knowledge is constructed based on the part fault characteristic frequency, which comprises the following steps:

[0059] 1.1) Calculating the part fault characteristic frequency based on the fault principle of the part and the parameter information of the part, and expanding the frequency to the left and right with the part fault characteristic frequency as the center to form a characteristic frequency band;

[0060] 1.2) Using the characteristic frequency band to perform band-pass filtering on the collected vibration data of the mechanical equipment to obtain the fault characteristic frequency, which is the guidance information of physical knowledge.

[0061] In this embodiment, for example, the failure feature frequency of the part is calculated by the pitch circle diameter of a certain type of bearing, the rolling element diameter, etc., and then, based on the calculated feature frequency, a feature frequency band is formed by extending a certain bandwidth, such as 5 Hz, on the left and right. The vibration information collected from the equipment is filtered using this feature frequency band to obtain the guidance information. Therefore, it is called physical knowledge-based guidance information, and it can also be called failure feature frequency guidance information.

[0062] In the above step 2), the generated failure feature frequency information is spliced with the failure data to form double-channel data. Specifically, after splicing the guidance information with the originally collected vibration data according to the different parts and needs, the spectrum conversion is performed to convert to the frequency domain or time-frequency domain to form double-channel data.

[0063] In this embodiment, the spectrum conversion can be envelope spectrum, Fourier spectrum, wavelet transform or short-time Fourier transform in time-frequency domain, etc.

[0064] Specifically, since the traditional fault diagnosis is mainly based on theoretical analysis, the features related to the failure are extracted according to the working principle and physical characteristics of the mechanical equipment. For rotating parts, the vibration frequency is an important characteristic parameter. For example, when a bearing fails, the vibration frequency will contain a feature frequency component related to the bearing failure, which can be used to judge the failure state of the bearing. Due to factors such as manufacturing errors, wear during equipment operation, slight fluctuations in rotation speed, etc., the feature frequency may have a slight deviation during actual detection. In order to improve the performance of the deep neural network model, effectively extract key features from the original failure data, and use these features for fault diagnosis and classification, the present application constructs physical knowledge guidance information based on the feature frequency of the part, which is used for model training.

[0065] The above step 2) is described in detail taking a bearing as an example:

[0066] (1) First, the feature frequency of the bearing is calculated based on the bearing failure principle, including the rolling element inner ring passing frequency (BPFI), the rolling element outer ring passing frequency (BPFO), the rolling element self-spinning frequency (BSF), and the cage frequency (FTF). The calculation formulas are as follows:

[0067]

[0068] In the formula, f BPFI is the rolling element inner ring passing frequency; n b is the number of rolling elements; f r is the rotation frequency; d b is the rolling element diameter; D c is the bearing pitch diameter (i.e. the diameter of the circle where the rolling element center is located); α is the contact angle; f BPFOis the rolling element outer ring passing frequency; f BSF is the rolling element spin frequency; f FTF is the retainer frequency.

[0069] (2) Based on the calculated characteristic frequency, each left and right is extended mHz to form a characteristic frequency band to solve the problem of slight deviation of the characteristic frequency. According to the needs of specific models and data sets, the k harmonic frequency bands are calculated, k = [1…N].

[0070] (3) The generated characteristic frequency band is used to band-pass filter the vibration information; and the vibration information and the information filtered using the characteristic band are converted into the frequency domain at the same time, according to the different parts, the bearing is converted into an envelope spectrum, and the gear, flywheel, etc. is converted into a Fourier spectrum.

[0071] As shown in Figure 3 , it is an envelope spectrum in a composite fault of a bearing outer ring and a flywheel imbalance. It can be seen that there are obvious imbalance fault characteristic frequencies, rotational frequencies and their second harmonics, and bearing outer ring fault characteristic frequencies, BPFO and its second harmonics. As shown in Figure 4 , it is the characteristic frequency information in a composite fault of a bearing outer ring and a flywheel imbalance, including bearing inner ring fault, outer ring fault, rolling element fault and 1-5 times frequency of rotational frequency. As can be seen from Figure 3 , Figure 4 , the main characteristic frequencies of the bearing outer ring fault and the flywheel imbalance fault are retained. These two information are input into the deep neural network as two channels, guiding the model to pay attention to the fault characteristics.

[0072] In the above step 2), the multi-output deep convolutional neural network adopts a parallel convolutional network, and the number of parallel convolutional networks is determined according to the part data or the type of spectrum signal.

[0073] In this embodiment, any conventional convolutional network can be used, which is not limited. The number of parallel convolutional networks can be determined according to the part data or the type of spectrum signal. For example, one convolutional network corresponds to each type of part, or one convolutional network corresponds to an envelope spectrum, and another convolutional network corresponds to an FFT.

[0074] In the above step 3), the multi-output classification aims to classify the samples into multiple outputs, and each output can be classification, regression or multi-label. The multi-output classification method realizes decoupling classification of composite faults through multiple output branches, and each output branch corresponds to a single fault type. Each branch uses a sigmoid activation function to realize prediction of a single fault type.

[0075] Specifically, as shown in Figure 5 , the multi-output deep convolutional neural network is used for multi-output classification of fault characteristics, including the following steps:

[0076] 3.1) Use the flattened layer to convert the extracted multi-dimensional composite fault features into one-dimensional features, or use the global average pooling layer for dimension reduction;

[0077] 3.2) Connect multiple output branches, each branch corresponding to a fault type; in each branch, select N layers of fully connected layers as needed, and finally output the probability of the fault by a fully connected layer with only one output and a sigmoid activation function, to determine whether the fault exists.

[0078] In the above embodiment, after the multi-output classification of the fault in step 3.2), there is a step 3.3) of constraint: directly splicing all the single fault probabilities of the outputs together and performing multi-output constraint to further constrain each independent branch output to achieve the setting of precision and loss function.

[0079] In this embodiment, the multi-output constraint is that the F1 score and the cross-entropy loss in the convolutional network are used to further constrain each independent branch output.

[0080] Specifically, in the multi-output fault diagnosis model, when each output branch independently predicts a specific type of fault, it may overfit the noise or background features in the training data, which can cause the model to perform poorly in a composite fault scenario. To solve this problem, all output branches of the model are spliced together in parallel, and a cross-entropy loss is used for constraint, which encourages the model to learn more general features and improves the robustness and reliability of the model in actual application, effectively preventing overfitting of single fault features during training.

[0081] In summary, the present application is based on the guidance information of the physical knowledge constructed based on the part feature frequency, which is used for model training to guide the model to learn the correct fault features; the multi-output mode in the deep neural network is used to realize the decomposition and diagnosis of the mechanical equipment composite fault, each output corresponds to a fault type of the mechanical equipment, the features corresponding to the fault are extracted from the global features, and the diagnosis of whether the fault exists in the equipment is completed, which can extract correct fault features and has good generalization ability; at the same time, it can realize accurate diagnosis of the running state of the mechanical equipment, and when there are multiple faults in the system, it can be decomposed and diagnosed as multiple individual faults.

[0082] In one embodiment of the present application, a fault diagnosis system based on part feature frequency is provided, which comprises:

[0083] The physical knowledge information acquisition module acquires the part fault feature frequency from the original fault data set of the mechanical equipment, and constructs the guidance information of the physical knowledge based on the part fault feature frequency;

[0084] The feature extraction module splices the generated fault feature frequency information with the fault data to form double-channel data, and inputs the double-channel data into a deep convolutional neural network with multiple outputs for training to extract fault features.

[0085] The fault decoupling classification and diagnosis module performs multi-output classification and diagnosis on the fault features by the deep convolutional neural network with multiple outputs, each output corresponding to a fault type of the mechanical equipment, extracts fault features corresponding to the fault type from the global features, and completes diagnosis of whether the fault exists in the mechanical equipment.

[0086] In the above embodiment, the physical knowledge information acquisition module acquires part fault feature frequencies from a mechanical equipment original fault data set, and constructs guidance information of physical knowledge based on the part fault feature frequencies, including:

[0087] Based on the fault principle of the part and the parameter information of the part, the part fault feature frequency is calculated, and the feature frequency band is formed by performing left and right spectrum spreading around the part fault feature frequency.

[0088] The feature frequency band is used for band-pass filtering of the collected vibration data of the mechanical equipment, and the obtained fault feature frequency is the guidance information of the physical knowledge.

[0089] In the embodiment, the generated fault feature frequency information is spliced with the fault data to form double-channel data, specifically: after splicing the guidance information with the originally collected vibration data according to the different parts and needs, frequency spectrum conversion is performed to convert to the frequency domain or time-frequency domain to form double-channel data.

[0090] In the above embodiment, the deep convolutional neural network with multiple outputs adopts a parallel convolutional network, and the number of parallel convolutional networks is determined according to the part data or the type of the frequency spectrum signal.

[0091] In the above embodiment, the fault decoupling classification and diagnosis module performs multi-output classification on the fault features by the deep convolutional neural network with multiple outputs, including:

[0092] The extracted multi-dimensional complex fault features are converted into one-dimensional features by using a flattening layer, or dimension reduction is performed by using a global average pooling layer.

[0093] A plurality of output branches are connected, each branch corresponding to a fault type; in each branch, N-layer fully connected layers are selected as needed, and finally a fully connected layer with only one output and using a sigmoid activation function outputs the probability of the fault to determine whether the fault exists.

[0094] In the embodiment, after multi-output classification of the fault, a constraint is further included: the single fault probabilities of all outputs are spliced together, multi-output constraints are performed, and each independent branch output is further constrained.

[0095] The multi-output constraint is that the F1 score in the convolutional network and the cross-entropy loss are used to further constrain each independent branch output.

[0096] The system provided by the embodiment is used for executing the above-mentioned method embodiments, and the specific process and detailed content are referred to the above-mentioned embodiments, which will not be described here.

[0097] In an embodiment of the present application, a computing device is provided, which can be a terminal, and can include a processor, a communications interface, a memory, a display screen and an input device. The processor, the communications interface and the memory complete mutual communication through a communication bus. The processor is used to provide computing and control capabilities. The memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program, which is executed by the processor to implement the method in each embodiment. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communications interface is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, a management network, NFC (near field communication) or other technologies. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computing device, or an external keyboard, touchpad or mouse, etc. The processor can call the logical instructions in the memory.

[0098] In addition, the logical instructions in the memory described above can be implemented in the form of a software functional unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0099] In an embodiment of the present application, a computer program product is provided, the computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions that, when executed by a computer, enable the computer to carry out the method provided by any of the above method embodiments.

[0100] In an embodiment of the present application, a non-transitory computer readable storage medium is provided, the non-transitory computer readable storage medium storing server instructions, the computer instructions causing a computer to perform the method provided by any of the above embodiments.

[0101] The computer readable storage medium provided by the above embodiments has similar implementation principles and technical effects to the above method embodiments, and thus will not be described here.

[0102] The present application is described in reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0103] These computer program instructions can also be stored in a computer readable memory that can direct the computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0104] These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide a process for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0105] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; and although the present application has been described in detail with reference to the foregoing embodiments, it should be appreciated by those skilled in the art that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features thereof can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A fault diagnosis method based on the characteristic frequency of a component, characterized in that, include: The fault characteristic frequencies of parts are obtained from the original fault dataset of mechanical equipment. Based on the fault characteristic frequencies of parts, guidance information for physical knowledge is constructed, including: calculating the fault characteristic frequencies of parts based on the fault principle and parameter information of parts; spreading the frequency band to the left and right with the fault characteristic frequencies of parts as the center to form a characteristic frequency band; using the characteristic frequency band to perform bandpass filtering on the vibration data of the collected mechanical equipment, and the obtained fault characteristic frequencies are the guidance information for physical knowledge. The generated fault feature frequency information is concatenated with the fault data to form dual-channel data, which is then input into a multi-output deep convolutional neural network for training to extract fault features. A multi-output deep convolutional neural network is used to classify and diagnose fault features. Each output corresponds to a fault type of mechanical equipment. Fault features corresponding to the fault type are extracted from global features to complete the diagnosis of whether the fault exists in the mechanical equipment.

2. The fault diagnosis method based on component characteristic frequencies as described in claim 1, characterized in that, The generated fault characteristic frequency information is concatenated with the fault data to form dual-channel data, including: Depending on the different parts and requirements, the guidance information is spliced ​​with the original collected vibration data, and then the spectrum is converted to the frequency domain or time-frequency domain to form dual-channel data.

3. The fault diagnosis method based on component characteristic frequencies as described in claim 1, characterized in that, Multi-output deep convolutional neural networks employ parallel convolutional networks, the number of which is determined based on the part data or the type of spectral signal.

4. The fault diagnosis method based on the characteristic frequency of a component as described in claim 1, characterized in that, Fault features are classified using a multi-output deep convolutional neural network, including: The extracted multidimensional composite fault features can be converted into one-dimensional features by using a flattening layer, or a global average pooling layer can be used for dimensionality reduction. Multiple output branches are connected, each branch corresponding to a fault type. In each branch, N fully connected layers are selected as needed. Finally, a fully connected layer with only one output and using the sigmoid activation function outputs the probability of the fault to determine whether the fault exists.

5. The fault diagnosis method based on the characteristic frequency of a component as described in claim 4, characterized in that, After classifying faults into multiple outputs, the process also includes a constraint step: The single-failure probabilities of all outputs are concatenated together to apply multi-output constraints, which further constrain the output of each independent branch.

6. The fault diagnosis method based on the characteristic frequency of a component as described in claim 5, characterized in that, The multi-output constraint is to further constrain the output of each independent branch using the F1 score and cross-entropy loss in the convolutional network.

7. A fault diagnosis system based on the characteristic frequency of a component, characterized in that, include: The physical knowledge information acquisition module obtains the component fault characteristic frequencies from the original fault dataset of mechanical equipment, and constructs physical knowledge guidance information based on the component fault characteristic frequencies. This includes: calculating the component fault characteristic frequencies based on the component's fault principle and parameter information, spreading the frequency band left and right around the component fault characteristic frequencies to form a characteristic frequency band; and using the characteristic frequency band to perform bandpass filtering on the collected vibration data of the mechanical equipment to obtain the fault characteristic frequencies as the physical knowledge guidance information. The feature extraction module concatenates the generated fault feature frequency information with the fault data to form dual-channel data, which is then input into a multi-output deep convolutional neural network for training to extract fault features. The fault decoupling classification and diagnosis module uses a multi-output deep convolutional neural network to perform multi-output classification and diagnosis of fault features. Each output corresponds to a fault type of mechanical equipment. The module extracts the fault features corresponding to the fault type from the global features to complete the diagnosis of whether the fault exists in the mechanical equipment.

8. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 6.

9. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 6.

Images