A method and system for unsupervised intelligent fault diagnosis of mechanical parts

Abstract

The present invention discloses a method and system for unsupervised intelligent fault diagnosis of mechanical parts, which relates to the field of intelligent fault diagnosis technology, including: preparing an unsupervised intelligent fault diagnosis data set for rotating machinery and building an unsupervised intelligent fault diagnosis model for rotating machinery; training the unsupervised intelligent fault diagnosis model and designing a loss function that comprehensively considers mapping-based and adversarial-based loss functions as optimization targets; evaluating the model, inputting test set samples, outputting prediction results, and obtaining the performance effect of the model when migrating source domain knowledge to the target domain under unsupervised conditions. The unsupervised intelligent fault diagnosis method for mechanical parts provided by the present invention focuses on optimizing the backbone network and domain adaptation learning method of the model, improving model performance, improving feature engineering, and comprehensively considering the reverse optimization process of classification loss, mapping domain migration loss, and adversarial domain migration loss, so as to improve the accuracy of the model, without the need for additional hardware facilities, and reducing the cost of use.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent fault diagnosis, and in particular to an unsupervised intelligent fault diagnosis method and system for mechanical parts. Background Art

[0002] The manufacturing industry has stringent requirements for production and processing accuracy, but rotating machinery will inevitably malfunction after long-term operation, affecting the quality of the processing process and posing the risk of equipment damage. Therefore, effective fault diagnosis and monitoring of the processing process is particularly important, namely, prognostics health management (PHM). Intelligent manufacturing methods and means that incorporate machine learning are leading the transformation of the manufacturing industry. Among them, intelligent fault diagnosis methods have become a research hotspot in the field of PHM. By using deep learning models (such as convolutional neural networks, recurrent neural networks, and autoencoders), these methods avoid the limitations of traditional machine learning methods (such as support vector machines and random forests) that manually extract time and frequency domain features. They adaptively obtain high-quality information from features in an end-to-end manner and output diagnostic results accurately and efficiently.

[0003] The effectiveness of deep learning in intelligent fault diagnosis is based on the following two assumptions: there is a large amount of labeled data; and the failure modes of the training dataset in the source domain are the same as the failure modes of the test dataset in the target domain (mathematically, the training dataset (source domain) should follow the same distribution as the test dataset (target domain)). However, the training datasets obtained in the laboratory are not completely consistent with the data generated in actual industrial equipment. If deep learning models are trained using these datasets, they may overfit the training datasets, resulting in weak generalization to actual industrial applications (especially for new conditions that were not trained in the model). In addition, in real applications, machines usually operate under various working conditions, which requires training models that can adapt to changes in working conditions. These two aspects limit the application scenarios of intelligent fault diagnosis and hinder the development of related methods.

[0004] Therefore, there is an urgent need for an unsupervised intelligent fault diagnosis method for mechanical parts to solve the problem of being unable to build a large-scale fault diagnosis data set to support the training needs of deep learning models, and to improve the feasibility and effectiveness of deep learning methods in the field of intelligent fault diagnosis. Summary of the Invention

[0005] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, the technical problem solved by the present invention is: when it is impossible to build a large-scale fault diagnosis data set to support the training needs of the deep learning model, the feasibility and effectiveness of the deep learning method in the field of intelligent fault diagnosis are improved.

[0008] To solve the above technical problems, the present invention provides the following technical solutions: an unsupervised intelligent fault diagnosis method for mechanical parts, comprising:

[0009] Create a rotating machinery unsupervised intelligent fault diagnosis dataset and build an unsupervised intelligent fault diagnosis model for rotating machinery;

[0010] Train an unsupervised intelligent fault diagnosis model and design a comprehensive loss function based on mapping and adversarial loss as the optimization objective;

[0011] Evaluate the model, input test set samples, output prediction results, and obtain the performance of the model when transferring source domain knowledge to the target domain under unsupervised conditions.

[0012] As a preferred solution of the unsupervised intelligent fault diagnosis method for mechanical parts described in the present invention, the fault diagnosis data set includes: dividing the data set into a source domain and a target domain, and dividing it into a training set and a test set.

[0013] As a preferred solution of the unsupervised intelligent fault diagnosis method for mechanical parts described in the present invention, the fault diagnosis model includes: a backbone feature extraction network and a domain adaptive migration algorithm.

[0014] As a preferred solution of the unsupervised intelligent fault diagnosis method for mechanical parts of the present invention, the backbone feature extraction network includes:

[0015] Based on the one-dimensional residual convolutional network, it integrates the hole convolution unit, the shrinkage denoising unit, and the non-local block unit;

[0016] The dilated convolution unit uses a convolution kernel with two expansion rates to extract features from the vibration signal, and determines whether the model can learn the multi-scale features of the vibration signal through the receptive field;

[0017] The shrinkage noise reduction unit includes automatically optimizing the noise reduction filter parameters used by the back propagation algorithm, eliminating noise information and extracting effective features by combining soft thresholding and deep learning;

[0018] The Non-LocalBlock unit includes constructing a non-local operation operator that has a simple structure and can be directly embedded in any deep network to capture the connection between time, space and long-range features of spatiotemporal data;

[0019] The domain adaptive migration algorithm includes PolyLoss-CE, which comprehensively considers the classification loss of source domain data, the joint maximum average difference, and the corresponding values ​​of the conditional domain adversarial network.

[0020] As a preferred solution of the unsupervised intelligent fault diagnosis method for mechanical parts of the present invention, the receptive field is expressed as:

[0021]

[0022] Among them, RF l+1 is the receptive field size of the current feature map; RF l is the receptive field size corresponding to the feature map of the previous layer; f l+1 Indicates the current convolution kernel size; s i is the step size of the i-th convolutional layer, and a is the expansion rate.

[0023] As a preferred solution of the unsupervised intelligent fault diagnosis method for mechanical parts of the present invention, the Non-LocalBlock unit is represented as:

[0024]

[0025] Where x is the input feature map, i and j are index values, the f function calculates the similarity between i and j, the g function calculates the corresponding representation of the feature map at position j, and C(x) is the response factor;

[0026] The loss function is expressed as,

[0027]

[0028] Among them, L CE is the standard cross entropy loss; ξ is the polynomial coefficient; y and They are respectively the real fault label and label classification network G y The predicted output of

[0029] JMMD focuses on the joint probability distribution of the output features and is expressed as:

[0030]

[0031] Among them, H k is the RKHS using kernel k (usually a Gaussian kernel); E is the mathematical expectation; x sand x t are the data distribution of the source domain and the target domain respectively; is the mapping of the feature tensor product in the Hilbert space; L is the specified high-dimensional feature output layer; z sl and z tl are the activations of the lth layer in the source and target domains respectively;

[0032] Let z be the feature space. The framework mainly consists of feature mapping network Label classification network G y (z;θ y ) and domain discrimination network G d (z;θ d ), the CDAN loss value is expressed as:

[0033]

[0034] w(H(p))=1+e -H(p)

[0035]

[0036] Among them, G f Map the data of the source domain and the target domain into a feature space; G y Predict the source domain label y and use the standard cross entropy loss, G d The source of the predicted feature, i.e., the source domain or the target domain. G d The purpose is to make the data of the source domain and the target domain distributed consistently in z. represents the outer product operation, p c is the domain label;

[0037] The loss function of the unsupervised intelligent fault diagnosis model is expressed as,

[0038] L UFD =L Poly-CE +λ1L JMMD +λ2L CDAN .

[0039] As a preferred embodiment of the unsupervised intelligent fault diagnosis method for mechanical parts of the present invention, the training of the unsupervised intelligent fault diagnosis model includes:

[0040] Input training set data as model feature learning samples, input test data as model evaluation and verification samples after each training epoch, and improve model parameters;

[0041] The evaluation of the prediction model includes using overall accuracy and average accuracy as relevant indicators;

[0042] Use the average precision and maximum precision indicators within the last 10 epochs;

[0043] The average accuracy and maximum accuracy indicators of the stage are used within 10 epochs after the model reaches the best performance.

[0044] Another object of the present invention is to provide an unsupervised intelligent fault diagnosis system for mechanical parts, which can solve the problem of how to reduce noise while maximizing the retention of original feature information by designing an end-to-end efficient feature extraction network.

[0045] To solve the above technical problems, the present invention provides the following technical solutions: an unsupervised intelligent fault diagnosis system for mechanical parts, comprising:

[0046] Data acquisition module, fault diagnosis module, control module, display module;

[0047] As a preferred solution of the unsupervised intelligent fault diagnosis system for mechanical parts of the present invention, the data acquisition module is a device for acquiring equipment data, which is used to collect vibration signals of rotating equipment at a fixed frequency and upload them to the fault diagnosis module after completion;

[0048] As a preferred solution of the unsupervised intelligent fault diagnosis system for mechanical parts of the present invention, the fault diagnosis module is a device for performing diagnosis through data, which is used to receive data uploaded by the data acquisition module, calculate the current fault status of the rotating equipment, and display it on the display module;

[0049] As a preferred solution of the unsupervised intelligent fault diagnosis system for mechanical parts of the present invention, the control alarm module is a device that issues an alarm based on the diagnostic information, and is used to receive the fault diagnosis information transmitted by the fault diagnosis module. When a fault occurs, the control alarm device is activated and a relevant prompt voice is played;

[0050] As a preferred solution of the unsupervised intelligent fault diagnosis system for mechanical parts described in the present invention, the display module is a device for displaying equipment fault conditions, and is used to connect the data acquisition module, fault diagnosis module, control module, and alarm module through a bus.

[0051] A computer device includes a memory and a processor, wherein the memory stores a computer program, and is characterized in that the processor implements the steps of the above method when executing the computer program.

[0052] A computer-readable storage medium stores a computer program thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0053] Beneficial effects of the present invention: The unsupervised intelligent fault diagnosis method for mechanical parts provided by the present invention is based on deep learning and domain adaptation theory. Under the premise of an incomplete data set (containing only source domain training samples), the feature learning ability is transferred, and the distribution difference between the source domain and the target domain is continuously reduced, thereby building a strong robust prediction model with domain-invariant characteristics based on the source domain data.

[0054] We focus on optimizing the model's backbone network and domain adaptation learning method, integrating the dilated convolution unit, shrinkage denoising unit, and non-local block unit to improve model performance from the perspectives of global feature receptive field, signal adaptive denoising, and suppressing long-range feature information loss, thereby improving feature engineering.

[0055] Furthermore, a comprehensive inverse optimization process that considers classification loss, mapping domain transfer loss, and adversarial domain transfer loss further enhances model accuracy. It can be run directly on personal computers and embedded devices without the need for additional hardware. Its simplicity reduces costs and makes it highly suitable for deployment and implementation in any environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0057] Figure 1 An overall flow chart of an unsupervised intelligent fault diagnosis method for mechanical parts provided by one embodiment of the present invention.

[0058] Figure 2 This is an overall structural diagram of an unsupervised intelligent fault diagnosis system for mechanical parts provided by the second embodiment of the present invention.

[0059] Figure 3 This is a diagram showing the overall comparative experimental results of an unsupervised intelligent fault diagnosis method for mechanical parts provided by the fourth embodiment of the present invention on the CWRU dataset;

[0060] Figure 4 This is a diagram showing the overall comparison results of the PU dataset for an unsupervised intelligent fault diagnosis method for mechanical parts provided by the fourth embodiment of the present invention;

[0061] Figure 5 This is a diagram showing the overall comparison results of an unsupervised intelligent fault diagnosis method for mechanical parts in a PHM dataset provided by the fourth embodiment of the present invention;

[0062] Figure 6 This is a diagram showing the overall comparison results of an unsupervised intelligent fault diagnosis method for mechanical parts provided by the fourth embodiment of the present invention on the JNU dataset. DETAILED DESCRIPTION

[0063] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0064] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0065] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0066] The present invention is described in detail with reference to schematic diagrams. For ease of illustration, cross-sectional views of device structures may be partially enlarged and not to scale when describing embodiments of the present invention. Furthermore, the schematic diagrams are merely illustrative and should not limit the scope of the present invention. Furthermore, in actual production, the three-dimensional dimensions of length, width, and depth should be included.

[0067] In the description of the present invention, it should be noted that the terms "upper, lower, inner, and outer" and other references to orientations or positional relationships are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first, second, or third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0068] In this disclosure, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be interpreted broadly. For example, they may refer to fixed, removable, or integral connections. They may also refer to mechanical, electrical, or direct connections, indirect connections through an intermediary, or internal communication between two components. Those skilled in the art will understand the specific meanings of these terms in this disclosure.

[0069] Example 1

[0070] Reference Figure 1 , which is an embodiment of the present invention, provides an unsupervised intelligent fault diagnosis method for mechanical parts, comprising:

[0071] Create a rotating machinery unsupervised intelligent fault diagnosis dataset and build an unsupervised intelligent fault diagnosis model for rotating machinery;

[0072] The fault diagnosis dataset includes:

[0073] Use internationally available domain datasets to compare and evaluate algorithm performance;

[0074] Each dataset is divided into source domain and target domain data according to the working conditions, and 80% of the source domain and target domain data are used as training sets, and the remaining 20% ​​are used as test sets;

[0075] During the training process, the data set will be verified using the test set after each iteration to understand whether overfitting occurs during the training process.

[0076] The fault diagnosis model includes: backbone feature extraction network and domain adaptive migration algorithm.

[0077] Train an unsupervised intelligent fault diagnosis model and design a comprehensive loss function based on mapping and adversarial loss as the optimization objective;

[0078] The backbone feature extraction network includes:

[0079] Based on the one-dimensional residual convolutional network, it integrates the hole convolution unit, the shrinkage denoising unit, and the non-local block unit;

[0080] The dilated convolution unit uses convolution kernels with two dilation rates to extract features from vibration signals, and uses the receptive field to determine whether the model can learn multi-scale features of vibration signals.

[0081] The Shrinkage noise reduction unit automatically optimizes the noise reduction filter parameters used through the back propagation algorithm, eliminates noise information and extracts effective features through a combination of soft thresholding and deep learning;

[0082] The Non-LocalBlock unit includes a non-local operation operator that is simple in structure and can be directly embedded into any deep network to capture the connection between time, space and long-range features of spatiotemporal data;

[0083] The domain adaptive transfer algorithm includes PolyLoss-CE, which comprehensively considers the classification loss of source domain data, the joint maximum average difference, and the corresponding values ​​of the conditional domain adversarial network.

[0084] In the backbone feature extraction network, a shrinkage denoising unit is designed, and the denoising filter parameters used are automatically optimized through the back propagation algorithm. The noise information is eliminated and effective features are extracted by combining soft thresholding and deep learning.

[0085] The Shrinkag unit first applies global average pooling to the feature map to obtain a one-dimensional feature vector; then, the vector is input into a two-layer fully connected network to obtain a scaling parameter, which is normalized by the Sigmoid function; finally, the scaling parameter is multiplied by the previous one-dimensional vector channel mean as a threshold.

[0086] When a certain number of Bottleneck filters are added to this filter, the noise-related features in the signal will be gradually reduced.

[0087] The receptive field is expressed as:

[0088]

[0089] Among them, RF l+1 is the receptive field size of the current feature map; RF l is the receptive field size corresponding to the feature map of the previous layer; f l+1 Indicates the current convolution kernel size; s i is the step size of the i-th convolutional layer. It can be seen that RF l+1 With RF l 、f l+1 and s i The dilated convolution cleverly increases the receptive field without increasing the number of parameters. Its core idea is to introduce a hyper-parameter called the "dilation rate," which defines the spacing between the actual values ​​in the convolution kernel, filling the gaps with zeros. The receptive field calculation formula incorporating this idea can be expressed as:

[0090]

[0091] Where a is the dilation rate. Changing the value of a will change the receptive field. Unlike pooling layers, which perform a similar function, dilated convolutions expand the receptive field without losing resolution and by preserving the relative spatial positions of pixels.

[0092] In the backbone feature extraction network, a Non-LocalBlock unit is designed. This module captures the connections between long-range features in temporal (one-dimensional time series signals), spatial (images), and spatiotemporal (video sequences) data by constructing a simple non-local operator that can be directly embedded in any deep network. This unit requires minimal computation to capture long-range features, ensuring consistency between the input and output scales during its operation, and possesses strong universality.

[0093] Non-LocalBlock unit, represented as:

[0094]

[0095] Where x is the input feature map, i and j are index values, the f function calculates the similarity between i and j, the g function calculates the corresponding representation of the feature map at position j, and C(x) is the response factor;

[0096] The model's optimization objective is a combination of the source domain data classification loss, the Joint Maximum Mean Discrepancy (JMMD), and the Conditional Domain Adversarial Network (CDAN) using mapping-based and adversarial methods. Unlike traditional cross-entropy loss, the source domain data classification loss here uses PolyLoss-CE. PolyLoss-CE designs cross-entropy as a linear combination of polynomial functions and approximates the loss function through Taylor expansion. This method has achieved superior results on multiple deep learning tasks compared to traditional loss functions.

[0097] The loss function is expressed as,

[0098]

[0099] Among them, L CE is the standard cross entropy loss; ξ is the polynomial coefficient; y and They are respectively the real fault label and label classification network G y The predicted output of

[0100] JMMD focuses on the joint probability distribution of the output features and is expressed as:

[0101]

[0102] Among them, H k is the RKHS using kernel k, usually a Gaussian kernel; E is the mathematical expectation; x s and x t are the data distribution of the source domain and the target domain respectively; is the mapping of the feature tensor product in the Hilbert space; L is the specified high-dimensional feature output layer; z sl and z tl are the activations of the lth layer in the source and target domains respectively;

[0103] Let z be the feature space. The framework mainly consists of feature mapping network Label classification network G y (z;θ y ) and domain discrimination network G d (z;θ d ), the CDAN loss value is expressed as:

[0104]

[0105] w(H(p))=1+e -H(p)

[0106]

[0107] Among them, G f Map the data of the source domain and the target domain into a feature space; G y Predict the source domain label y and use the standard cross entropy loss, G d The source of the predicted feature, that is, the source domain or the target domain. G d The purpose is to make the data of the source domain and the target domain distributed consistently in z. represents the outer product operation, p c is the domain label;

[0108] The loss function of the unsupervised intelligent fault diagnosis model is expressed as,

[0109] L UFD =L Poly-CE +λ1L JMMD +λ2L CDAN .

[0110] Training an unsupervised intelligent fault diagnosis model involves:

[0111] Input training set data as model feature learning samples, input test data as model evaluation and verification samples after each training epoch, and improve model parameters;

[0112] Evaluate the model, input test set samples, output prediction results, and obtain the performance of the model when transferring source domain knowledge to the target domain under unsupervised conditions.

[0113] Evaluating the prediction model includes,using overall accuracy and average accuracy as relevant metrics;

[0114] Use the average precision and maximum precision indicators within the last 10 epochs;

[0115] The average accuracy and maximum accuracy indicators of the stage are used within 10 epochs after the model reaches the best performance.

[0116] Example 2

[0117] Reference Figure 2 , which is an embodiment of the present invention, provides an unsupervised intelligent fault diagnosis system for mechanical parts, comprising:

[0118] Data acquisition module 100, fault diagnosis module 200, control module 300, display module 400;

[0119] The data acquisition module 100 is a device for acquiring equipment data, which is used to collect vibration signals of rotating equipment at a fixed frequency and upload them to the fault diagnosis module 200 after completion;

[0120] The fault diagnosis module 200 is a device for performing diagnosis through data, which is used to receive the data uploaded by the data acquisition module 100, calculate the current fault condition of the rotating equipment and display it on the display module 400;

[0121] The control alarm module 300 is a device that issues an alarm based on the diagnostic information. It is used to receive the fault diagnosis information transmitted by the fault diagnosis module 200. When a fault occurs, it activates the alarm device and plays the relevant prompt voice.

[0122] The display module 500 is a device for displaying equipment fault conditions, and is used to connect the data acquisition module 100, the fault diagnosis module 200, the control module 300, and the alarm module 400 via a bus.

[0123] Example 3

[0124] An embodiment of the present invention is different from the previous two embodiments in that:

[0125] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling 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 method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0126] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0127] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0128] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0129] Example 4

[0130] Reference Figure 3-6 , which is an embodiment of the present invention, provides an unsupervised intelligent fault diagnosis method for mechanical parts. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0131] At present, most studies use supervised learning methods to achieve intelligent fault diagnosis, which has the following problems:

[0132] Vibration signals collected from the workspace of a machining center often contain significant noise. Directly inputting these noisy signals into a deep neural network can significantly reduce the model's ability to learn features, negatively impacting prediction accuracy. While noisy signals are typically preprocessed, this undoubtedly increases the uncertainty of this preprocessing effort. Designing an efficient, end-to-end feature extraction network that minimizes the impact of noise while preserving the original information is crucial.

[0133] Flexible production processes involve numerous variables, including inconsistent parts types, raw materials, operating conditions, and equipment types. Obtaining a comprehensive dataset for every operating condition for deep learning model training is difficult, if not impossible.

[0134] In this embodiment, a specific simulation experiment is conducted on the method of the present invention. Under a preset equivalent experimental environment, the algorithm of the above embodiment is compared with some mainstream deep learning power load forecasting structures to measure the economic benefits of the algorithm. The experimental results are as follows: Figure 3-4 As shown:

[0135] Specific experimental results such as Figure 3-6 As shown in the figure, ResNet18, ResNet50, TCN with denoising attention mechanism, CNN+BiLSTM+ECA_Attention, CNN+BiGRU+Coord_Attention, and the model proposed in this paper are named A, B, C, D, E, and F respectively.

[0136] The results show the local and global transfer effects of each model in the comparative experiments on different datasets. Overall, model F has the best comprehensive performance on the four public datasets, and is only slightly inferior to model A in CWRU.

[0137] The performance of F compared to other models on the four datasets can be summarized as follows:

[0138] In the CWRU dataset, the overall improvement ranges of Best-Max, Best-Mean, Last-Max, and Last-Mean are 0%-1.06%, 0.03%-4.46%, -0.64%-4.44%, and -0.26%-4.68%;

[0139] In the PU dataset, the overall improvement ranges of the four indicators are 10.02%-32.63%, 10.89%-33.31%, 9.15%-32.20%, and 9.26%-34.91%;

[0140] In the PHM dataset, the overall improvement ranges of the four indicators are 4.87%-19.00%, 3.99%-19.92%, 5.02%-21.58%, and 5.13%-21.54%; in the JNU dataset, the overall improvement ranges of the four indicators are 0.57%-2.59%, 0.40%-3.24%, 0.29%-2.59%, and 0.65%-2.70%.

[0141] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An unsupervised intelligent fault diagnosis method for mechanical parts, characterized in that: include: Create a rotating machinery unsupervised intelligent fault diagnosis dataset and build an unsupervised intelligent fault diagnosis model for rotating machinery; Train an unsupervised intelligent fault diagnosis model and design a comprehensive loss function based on mapping and adversarial loss as the optimization objective; Evaluate the model, input test set samples, output prediction results, and obtain the performance of the model when transferring source domain knowledge to the target domain under unsupervised conditions; The fault diagnosis model includes: a backbone feature extraction network and a domain adaptive migration algorithm; The backbone feature extraction network includes: based on a one-dimensional residual convolutional network, integrating a dilated convolution unit, a shrinkage noise reduction unit, and a non-local block unit; The dilated convolution unit uses a convolution kernel with two expansion rates to extract features from the vibration signal, and determines whether the model can learn the multi-scale features of the vibration signal through the receptive field; The shrinkage noise reduction unit includes automatically optimizing the noise reduction filter parameters used by the back propagation algorithm, eliminating noise information and extracting effective features by combining soft thresholding and deep learning; The Non-LocalBlock unit includes constructing a non-local operation operator to capture the connection between time, space and long-range features of spatiotemporal data; The domain adaptive migration algorithm includes PolyLoss-CE, which comprehensively considers the classification loss of source domain data, the joint maximum average difference, and the corresponding values ​​of the conditional domain adversarial network.

2. The unsupervised intelligent fault diagnosis method for mechanical parts according to claim 1, characterized in that: The fault diagnosis data set includes: The dataset is divided into source and target domains, and then into training and testing sets.

3. The unsupervised intelligent fault diagnosis method for mechanical parts according to claim 1 or 2, characterized in that: The receptive field is expressed as: Among them, RF l+1 is the receptive field size of the current feature map; RF l is the receptive field size corresponding to the feature map of the previous layer; f l+1 Indicates the current convolution kernel size; s i is the step size of the i-th convolutional layer, and a is the expansion rate.

4. The unsupervised intelligent fault diagnosis method for mechanical parts according to claim 3, characterized in that: The Non-LocalBlock unit is represented as: Among them, x is the input feature map, i and j are index values, the f function calculates the similarity between i and j, and the g function calculates the corresponding representation of the feature map at position j; The loss function is expressed as, Among them, L CE is the standard cross entropy loss; ξ is the polynomial coefficient; y and They are respectively the real fault label and label classification network G y The predicted output of JMMD focuses on the joint probability distribution of the output features and is expressed as: Among them, H k is the RKHS using kernel k; E is the mathematical expectation; x s and x t are the data distribution of the source domain and the target domain respectively; is the mapping of the feature tensor product in the Hilbert space; L is the specified high-dimensional feature output layer; z sl and z tl are the activations of the lth layer in the source and target domains respectively; Let z be the feature space, and the framework is mainly composed of feature mapping network Label classification network G y (z;θ y ) and domain discrimination network G d (z;θ d ), the CDAN loss value is expressed as: w(H(p))=1+e -H(p) Among them, G f Map the data of the source domain and the target domain into a feature space; G c Predict the source domain label, G d The source of the predicted feature, i.e. the source domain or the target domain, G d Make the data of source domain and target domain distributed consistently in z, represents the outer product operation, p s 、p t are the source domain and target domain labels; The loss function of the unsupervised intelligent fault diagnosis model is expressed as, L UFD =L Poly-CE +λ1L JMMD +λ2L CDAN 。 5. The unsupervised intelligent fault diagnosis method for mechanical parts according to claim 4, characterized in that: The training of the unsupervised intelligent fault diagnosis model includes: Input training set data as model feature learning samples, input test data as model evaluation and verification samples after each training epoch, and improve model parameters; The overall accuracy and average accuracy were used as relevant metrics to evaluate the model; Use the average precision and maximum precision indicators within the last 10 epochs; The average accuracy and maximum accuracy indicators of the stage are used within 10 epochs after the model reaches the best performance.

6. An unsupervised intelligent fault diagnosis system for mechanical parts, applying the unsupervised intelligent fault diagnosis method for mechanical parts according to any one of claims 1 to 5, characterized in that: include: Data acquisition module, fault diagnosis module, control module, display module; The data acquisition module is a device that obtains equipment data and is used to collect vibration signals of rotating equipment at a fixed frequency and upload them to the fault diagnosis module after completion; The fault diagnosis module is a device for performing diagnosis through data, which is used to receive the data uploaded by the data acquisition module, calculate the current fault condition of the rotating equipment and display it on the display module; The control module is a device that issues an alarm based on diagnostic information. It is used to receive the fault diagnosis information transmitted by the fault diagnosis module. When a fault occurs, it activates the alarm device and plays the relevant prompt voice. The display module is a device for displaying equipment fault conditions, and is used to connect the data acquisition module, the fault diagnosis module, the control module, and the alarm module through a bus.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the unsupervised intelligent fault diagnosis method for mechanical parts according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the unsupervised intelligent fault diagnosis method for mechanical parts according to any one of claims 1 to 5 are implemented.

Images