An open set assumption based intelligent fault diagnosis method for construction machinery

Abstract

The application discloses an engineering machine intelligent fault diagnosis method based on an open set assumption, and the method comprises the following steps: inputting labeled training data into a CNN model for training, and extracting activation characteristic values of the penultimate layer of each type of positive training sample; in order to evaluate the relationship between an input sample and each type of MAV, a combination of normalized Euclidean distance and cosine distance is adopted for distance calculation; the application solves the problem that, when test data appears in a fault type which is not contained in training data, the test data is not simply classified into a known type in the training set, but unknown type samples are accurately detected on the premise that the known type samples are correctly identified during fault diagnosis based on deep learning.

Description

Technical Field

[0001] This invention relates to an intelligent fault diagnosis method for engineering machinery based on the open set hypothesis, belonging to the field of bearing fault diagnosis technology. Background Technology

[0002] Health monitoring and fault diagnosis play a crucial role in ensuring the safety and reliability of modern machinery. The increasing complexity of machines and processes leads to various types of failures, such as lubrication contamination, leaks, overloads, and misalignments. These failures result in component degradation and overall machine failure, even dangerous malfunctions, causing production losses and personal injury. To improve mechanical reliability, detecting and identifying any type of early failure is essential for timely implementation of maintenance or control strategies. Previously, data-driven fault diagnosis attracted considerable attention due to its ability to learn the relationship between ever-growing monitoring data and machine health status. Among various data-driven diagnostic methods, deep learning-based fault diagnosis methods have been extensively studied, achieving accurate diagnostic predictions in an end-to-end manner, thus overcoming the drawbacks of traditional methods that require substantial expert knowledge and human intervention. The most successful deep learning models include stacked autoencoders, convolutional neural networks, and graph neural networks that mine hidden topological information. However, most current intelligent fault diagnosis models are based on deep learning methods with closed sets, meaning that the fault types in the training and testing data have the same categorical characteristics. Although methods under this assumption exhibit superior fault detection performance in the field of fault diagnosis. However, this assumption is largely invalid in industrial scenarios because the training set cannot encompass all fault types, thus posing significant limitations in practical industrial applications. On one hand, manually labeling all fault types during the training phase is costly and error-prone. On the other hand, vibration signal acquisition is time-consuming, making it difficult to collect data on all types of faults within a limited timeframe. Furthermore, some types of faults may not even occur. Deep learning models based on closed-set sums may misclassify fault types not previously seen in the test set as fault types defined in the training set. Therefore, considering real-world industrial machinery applications, the training and test sets are inconsistent. Bearing faults can occur on the inner ring, outer ring, balls, and cage, but the symptoms (such as pitting, wear, and fracture) and magnitude can differ. Consequently, the fault types during the testing phase are unpredictable and may include one or more unknown fault modes. Therefore, fault diagnosis models need the ability to detect unknown classes. To address this issue, the field of fault diagnosis needs to develop a method based on the open set hypothesis. This method can not only correctly classify known classes but also identify unknown classes. However, most current methods determine unknown classes by comparing thresholds with known classes. To more accurately determine the probability of each sample's class, we not only output the probability value corresponding to each known class but also calculate the probability value corresponding to the unknown class. Summary of the Invention

[0003] To address the problems existing in the prior art, this invention provides an intelligent fault diagnosis method for engineering machinery based on the open set hypothesis. This method solves the problem that when fault diagnosis based on deep learning occurs, if the test data contains fault types not included in the training data, it does not simply classify them as known classes in the training set. Instead, it accurately detects unknown class samples while correctly identifying known class samples.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] Firstly, this application provides an intelligent fault diagnosis method for engineering machinery based on the open set assumption, comprising the following steps:

[0006] Its training phase:

[0007] Step 1:

[0008] The labeled training data is input into the CNN model for training, and the activation feature value, or AV, of the penultimate layer of each class of positive training samples is extracted.

[0009] Let AV i,c Let m represent the AV value corresponding to the i-th positive training sample in the training set (c = 1, ..., K); the mean activation vector (MAV) corresponding to each known class is calculated using the nearest neighbor mean method, denoted as m. c m c It can be represented as:

[0010] m c =mean(AV) i,c ),

[0011] Step Two:

[0012] To evaluate the relationship between the input samples and the MAV of each class, d i,c Let x represent the i-th positive training sample. i and corresponding m c The distance between them; the distance is calculated using a combination of normalized Euclidean distance and cosine distance.

[0013] d i，c It can be represented as:

[0014] d i，c =||AV i，c -m c ||,

[0015] Step 3:

[0016] The distribution of each fault category is estimated based on the EVT method;

[0017] Since the EVT model follows a Weibull distribution, the Fithigh fitting function is used to fit the training samples and the corresponding m. c The maximum distance between them is used for Weibull fitting to obtain the Weibull model parameters for each known class; the Weibull fitting parameters ρ c Including displacement τ c Shape κ c and scale λ c This is used to estimate the probability of abnormal samples during the testing phase; as shown in the formula, the tail length η r This represents the selection of r maximum distances between the training samples and the MAV of each class;

[0018] ρ c =(τ c , κ c , λ c = FitHigh(||AV) i，c -m c ||,η r )

[0019] Next, the Weibull cumulative distribution function, or CDF probability, is prepared for the next process to estimate the class of the test sample; this is done by fitting the parameter ρ. c and distance d i，c Estimate the corresponding CDF i，c It evaluates the relationship between the i-th positive training sample and the c-th class; the formula is shown below:

[0020]

[0021] The known sample distribution is a monotonically increasing function, and the CDF value increases with the viewing distance.

[0022] Its testing phase:

[0023] S1: Input the preprocessed test data into the training model, where the activation feature value of the j-th test sample is represented as AV. j ;d j，c This represents the j-th test sample and the c-th known class m. c The distance between them is used to estimate whether the input is "far" from the known training data; d j，c It can be represented as:

[0024] d j，c =||m c -AV j ||,

[0025] S2: To assess the probability that a test sample is an outlier, CDF is used. j，cThe value of d is used to determine whether the j-th test sample belongs to the c-th known class; specifically, for the j-th test sample, the distance d j，c and the corresponding fitting parameters ρ c Used to calculate CDF j，c It measures the weights of input samples that do not belong to class c and is used as the core for estimating unknown class rejection; through distance d j，c Distance to the center of each feature class and fitting parameter ρ c The calculated values ​​are concentrated around 1, indicating that it does not belong to any known class. Therefore, when the test sample is of an unknown class, its corresponding CDF value is distributed around 1. Since known classes are included in the test sample, there are K+1 classes in the test sample. The K+1th class represents the unknown class sample, but the CNN model only has the output probabilities of K known classes, i.e., K class nodes, and lacks the K+1th class node. In order to calculate the output probability corresponding to the input sample belonging to the K+1 class nodes, the OpenMax method is used to correct the AV of the input sample, i.e., the AV value corresponding to the K-dimensional class node. Furthermore, the pseudo-AV of the unknown class corresponding to each sample is calculated, i.e., the (K+1)th class node.

[0026]

[0027]

[0028] OpenMax method for correcting AV values ​​using AV j and 1-CDF j The corresponding element-wise product is used to calculate the corrected AV. Therefore, when CDF j When it approaches 1, the corresponding known class correction for class c is... A value close to 0 indicates that the input test sample is more likely to be of the unknown class, and the expected pseudo-AV of the unknown class is the corresponding The value is greater;

[0029] However, due to the characteristics of vibration signals, especially the very clear data provided by Case Western Reserve University (CRWU), the AV distribution of class nodes for known class samples is relatively strong, all being greater than 0. However, due to the lack of prior knowledge about unknown class samples, the AV extraction values ​​for class nodes of unknown class samples show a negative distribution. In this case, the OpenMax method is continued to be used to correct the AV. When the sample is of an unknown class, the AV corresponding to the class node of that sample... j，i The values ​​are distributed negatively, while the corresponding CDF values ​​are... j，i As shown in the above formula, this will result in a pseudo-AV value for the unknown class. It is negative; however, the corrected AV value of the known class is AV. jThe value is close to 0, which causes the pseudo AV of the unknown class to be less than the corrected AV of the known class. This is inconsistent with the expectation, so it is easy to misclassify the known class as the unknown class.

[0030] Therefore, for the bearing fault dataset, the AV value is corrected by using an exponential function, and the pseudo AV value of the unknown class is further calculated to ensure the monotonicity of the extracted AV distribution and make it higher than the positive value.

[0031] The improved correction formula for AV is expressed as follows:

[0032]

[0033]

[0034] S3:

[0035] After obtaining the corrected AV, the probability value corresponding to each class is calculated to determine the class of each sample. However, the unknown class appears in the testing phase of the open set scenario; therefore, it contradicts the probability summation of 1. To calculate the probability distribution of each test sample belonging to a known class and the unknown class, the corrected average and the pseudo-average of the known class are used as the numerator, and their sum is used as the denominator, satisfying the probability summation requirement; therefore, the ability to reject input is provided when the unknown class, i.e., y = K + 1, has the highest probability.

[0036]

[0037] Secondly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of an intelligent fault diagnosis method for engineering machinery based on the open set hypothesis.

[0038] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of an intelligent fault diagnosis method for engineering machinery based on the open set hypothesis.

[0039] As can be seen from the above description, the present invention solves the problem that when fault types not included in the training data appear in the test data during deep learning-based fault diagnosis, it does not simply classify them as known classes in the training set, but accurately detects unknown class samples on the premise of correctly identifying known class samples. Attached Figure Description

[0040] Figure 1 This is a flowchart of an intelligent fault diagnosis method for engineering machinery based on the open set assumption of the present invention.

[0041] Figure 2This is the architecture diagram of the 1DCNN model of this invention; (the FC2 layer represents K-class nodes, and the Sigmoid layer represents the K-class functions corresponding to the K KCs)

[0042] Figure 3 A schematic diagram of the open set hypothesis fault diagnosis method during the training phase;

[0043] Figure 4 A schematic diagram of the open set hypothesis fault diagnosis method during the testing phase;

[0044] Figure 5 This is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments; however, it should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the scope of the present invention.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0047] An embodiment of the present invention provides a specific implementation of an intelligent fault diagnosis method for engineering machinery based on the open set assumption. See [link to implementation details]. Figure 1 The method specifically includes the following:

[0048] (1) Prepare the dataset:

[0049] training sample set Where x i Let y represent the i-th bearing sample. i ∈{C1, C2, ..., C M} represents sample x i The bearing fault category to which it belongs, where M represents the number of known categories contained in the bearing data of the training samples. Test sample set. The test set includes fault categories that do not belong to the training set, i.e., unknown bearing fault categories, which are classified as the K+1th category.

[0050] (2) Dataset processing method:

[0051] Data preparation: PHM2009 bearing dataset

[0052] First, the acquired bearing time-domain vibration data is converted to the frequency domain using a Fast Fourier Transform (FFT). This is primarily because frequency domain information is generally more sensitive to the machine's health status and typically provides better fault diagnosis performance. For the PHM2009 dataset, the vibration signals are first divided into data segments, with 6144 sampling points selected as each segment. Then, a FFT is performed on each signal, using 4097 coefficients in each sample. This process transforms each time-domain dataset into 100x4097-dimensional frequency domain data.

[0053] (3) Specific steps:

[0054] Convolutional Neural Network Architecture:

[0055] Compared to traditional machine learning methods, CNNs possess stronger learning capabilities and interpretability, and have been widely applied in intelligent fault diagnosis of mechanical parts. The time-domain or frequency-domain signals obtained after Fast Fourier Transform (FFT) can be directly used as input to CNNs, eliminating the need for manual feature extraction to obtain more advanced results. This reduces the time and effort required for feature extraction, meeting the real-time requirements of bearing fault diagnosis. However, current CNN methods are primarily used for closed-set assumption problems. To address the issue of unpredictable fault categories during testing, an open-set fault diagnosis model based on CNNs is proposed.

[0056] The CNN architecture comprises six convolutional layers and pooling layers. To suppress the impact of high-frequency noise on the classification accuracy of bearing signals, the first layer uses wide convolutional kernels. Specifically, the first convolutional kernel is 64×1 in size, and the remaining kernels are 3×1 in size. To standardize the data, a batch normalization layer is added after each convolutional layer. Two fully connected layers follow the convolutional layers; the nodes in the second fully connected layer correspond to the fault types in the training set. The final layer of the model consists of K one-to-many classifiers constructed using K sigmoid activation functions; that is, K sigmoid functions are used instead of the traditional softmax function. Specifically, K sigmoid functions are created for the K known classes. For N in the training set... train x training samples, each bearing sample x i They all have a class tag y i It contains K known classes, and let the l-th Sigmoid function correspond to the fault category C. l C is about to be l All corresponding samples are taken as positive examples, representing that these samples belong to fault class C. l All other known fault classes are considered as counterexamples, representing samples that do not belong to other known bearing fault classes, where l belongs to K, and the log loss function is Loss. l as follows:

[0057]

[0058] Where II represents the indicator function, that is, when the input is True, the output is 1, and when the input is False, the output is 0. l This represents the probability output of the l-th Sigmoid function for the i-th class;

[0059] The sum of the logarithmic loss functions of all sigmoid functions is used as the cost function to train the model, that is:

[0060]

[0061] Training phase:

[0062] Step 1:

[0063] The labeled training data is input into the CNN model for training, and the activation feature value, or AV, of the penultimate layer of each class of positive training samples is extracted.

[0064] Let AV i，c Let m represent the AV value corresponding to the i-th positive training sample in the training set (c = 1, ..., K); the mean activation vector (MAV) corresponding to each known class is calculated using the nearest neighbor mean method, denoted as m. c m c It can be represented as:

[0065] m c =mean(AV) i，c ),

[0066] Step Two:

[0067] To evaluate the relationship between the input samples and the MAV of each class, d i，c Let x represent the i-th positive training sample. i and corresponding m c The distance between them; the distance is calculated using a combination of normalized Euclidean distance and cosine distance.

[0068] d i，c It can be represented as:

[0069] d i，c =||AV i，c -m c ||,

[0070] Step 3:

[0071] The distribution of each fault category is estimated based on the EVT method;

[0072] Since the EVT model follows a Weibull distribution, the Fithigh fitting function is used to fit the training samples and the corresponding m.c The maximum distance between them is used for Weibull fitting to obtain the Weibull model parameters for each known class; the Weibull fitting parameters ρ c Including displacement τ c Shape κ c and scale λ c This is used to estimate the probability of abnormal samples during the testing phase; as shown in the formula, the tail length η r This represents the selection of r maximum distances between the training samples and the MAV of each class;

[0073] ρ c =(τ c , κ c , λ c = FitHigh(||AV) i，c -m c ||,η r )

[0074] Next, the Weibull cumulative distribution function, or CDF probability, is prepared for the next process to estimate the class of the test sample; this is done by fitting the parameter ρ. c and distance d i，c Estimate the corresponding CDF i，c， It evaluates the relationship between the i-th positive training sample and the c-th class; the formula is shown below:

[0075]

[0076] The known sample distribution is a monotonically increasing function, and the CDF value increases with the viewing distance.

[0077] Testing phase:

[0078] S1: Input the preprocessed test data into the training model, where the activation feature value of the j-th test sample is represented as AV. j ;d j，c This represents the j-th test sample and the c-th known class m. c The distance between them is used to estimate whether the input is "far" from the known training data; d j，c It can be represented as:

[0079] d j，c =||m c -AV j ||,

[0080] S2: To assess the probability that a test sample is an outlier, CDF is used. j，c The value of d is used to determine whether the j-th test sample belongs to the c-th known class; specifically, for the j-th test sample, the distance d j，c and the corresponding fitting parameters ρ cUsed to calculate CDF j，c It measures the weights of input samples that do not belong to class c and is used as the core for estimating unknown class rejection; through distance d j，c Distance to the center of each feature class and fitting parameter ρ c The calculated values ​​are concentrated around 1, indicating that it does not belong to any known class. Therefore, when the test sample is of an unknown class, its corresponding CDF value is distributed around 1. Since known classes are included in the test sample, there are K+1 classes in the test sample. The K+1th class represents the unknown class sample, but the CNN model only has the output probabilities of K known classes, i.e., K class nodes, and lacks the K+1th class node. In order to calculate the output probability corresponding to the input sample belonging to the K+1 class nodes, the OpenMax method is used to correct the AV of the input sample, i.e., the AV value corresponding to the K-dimensional class node. Furthermore, the pseudo-AV of the unknown class corresponding to each sample is calculated, i.e., the (K+1)th class node.

[0081]

[0082]

[0083] OpenMax method for correcting AV values ​​using AV j and 1-CDF j The corresponding element-wise product is used to calculate the corrected AV. Therefore, when CDF j When it approaches 1, the corresponding known class correction for class c is... A value close to 0 indicates that the input test sample is more likely to be of the unknown class, and the expected pseudo-AV of the unknown class is the corresponding The value is greater;

[0084] However, due to the characteristics of vibration signals, especially the very clear data provided by Case Western Reserve University (CRWU), the AV distribution of class nodes for known class samples is relatively strong, all being greater than 0. However, due to the lack of prior knowledge about unknown class samples, the AV extraction values ​​for class nodes of unknown class samples show a negative distribution. In this case, the OpenMax method is continued to be used to correct the AV. When the sample is of an unknown class, the AV corresponding to the class node of that sample... j，i The values ​​are distributed negatively, while the corresponding CDF values ​​are... j，i As shown in the above formula, this will result in a pseudo-AV value for the unknown class. It is negative; however, the corrected AV value of the known class is AV. j The value is close to 0, which causes the pseudo AV of the unknown class to be less than the corrected AV of the known class. This is inconsistent with the expectation, so it is easy to misclassify the known class as the unknown class.

[0085] Therefore, for the bearing fault dataset, the AV value is corrected by using an exponential function, and the pseudo AV value of the unknown class is further calculated to ensure the monotonicity of the extracted AV distribution and make it higher than the positive value.

[0086] The improved correction formula for AV is expressed as follows:

[0087]

[0088]

[0089] S3:

[0090] After obtaining the corrected AV, the probability value corresponding to each class is calculated to determine the class of each sample. However, the unknown class appears in the testing phase of the open set scenario; therefore, it contradicts the probability summation of 1. To calculate the probability distribution of each test sample belonging to a known class and the unknown class, the corrected average and the pseudo-average of the known class are used as the numerator, and their sum is used as the denominator, satisfying the probability summation requirement; therefore, the ability to reject input is provided when the unknown class, i.e., y = K + 1, has the highest probability.

[0091]

[0092] As can be seen from the above description, the present invention solves the problem that when fault types not included in the training data appear in the test data during deep learning-based fault diagnosis, it does not simply classify them as known classes in the training set, but accurately detects unknown class samples on the premise of correctly identifying known class samples.

[0093] Embodiments of this application also provide a specific implementation of an electronic device capable of implementing all steps in the above-described intelligent fault diagnosis method for engineering machinery based on the open set assumption. See [link to implementation details]. Figure 5 The electronic devices specifically include the following:

[0094] Processor, memory, communications interface, and bus;

[0095] The processor, memory, and communication interface communicate with each other via a bus.

[0096] The processor is used to invoke a computer program stored in memory. When the processor executes the computer program, it implements all the steps in the above embodiments.

[0097] As can be seen from the above description, the present invention solves the problem that when fault types not included in the training data appear in the test data during deep learning-based fault diagnosis, it does not simply classify them as known classes in the training set, but accurately detects unknown class samples on the premise of correctly identifying known class samples.

[0098] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the intelligent fault diagnosis method for engineering machinery based on the open set assumption described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the intelligent fault diagnosis method for engineering machinery based on the open set assumption described in the above embodiments.

[0099] As can be seen from the above description, the present invention solves the problem that when fault types not included in the training data appear in the test data during deep learning-based fault diagnosis, it does not simply classify them as known classes in the training set, but accurately detects unknown class samples on the premise of correctly identifying known class samples.

[0100] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, hardware + program embodiments are relatively simple in description because they are fundamentally similar to method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0101] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0102] While this application provides the method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the methods shown in the embodiments or drawings can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0103] While this specification provides method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive means. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or end product execution, the methods shown in the embodiments or drawings may be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed data processing environment). The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, the presence of other identical or equivalent elements in the process, method, product, or apparatus that includes said elements is not excluded.

[0104] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing the embodiments of this specification, the functions of each module can be implemented in one or more software and / or hardware components, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0105] Those skilled in the art will also know that, besides implementing the controller using purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the controller function as logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers (PLCs), and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices within it used to implement various functions can also be considered structures within that hardware component. Alternatively, the devices used to implement various functions can be considered as both software modules implementing the method and structures within a hardware component.

[0106] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0107] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0108] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0109] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0110] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0111] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0112] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of computer program products implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0113] The embodiments described in this specification can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. The embodiments of this specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0114] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, system embodiments are basically similar to method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. In the description of this specification, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments in this specification. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0115] The above description is merely an embodiment of the present specification and is not intended to limit the embodiments of the present specification. For those skilled in the art, various modifications and variations can be made to the embodiments of the present specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the embodiments of the present specification should be included within the scope of the claims of the embodiments of the present specification.

Claims

1. A method for intelligent fault diagnosis of engineering machinery based on the open set assumption, characterized by the following specific steps: Its training phase: Step 1: The labeled training data is input into the CNN model for training, and the activation feature value, or AV, of the penultimate layer of each class of positive training samples is extracted. Let AV i,c Let represent the AV value corresponding to the i-th positive training sample in the training set, where c = 1, ..., K; The mean activation vector (MAV) corresponding to each known class is calculated using the nearest neighbor mean method, denoted as m. c m c It can be represented as: m c =mean(OFF i,c ), Step Two: To evaluate the relationship between the input samples and the MAV of each class, d i,c Let x represent the i-th positive training sample. i and corresponding m c The distance between them; the distance is calculated using a combination of normalized Euclidean distance and cosine distance. d i,c It can be represented as: d i,c =||OFF i,c - m c ||, Step 3: The distribution of each fault category is estimated based on the EVT method; Since the EVT model follows a Weibull distribution, the Fithigh fitting function is used to fit the training samples and the corresponding m. c The maximum distance between them is used for Weibull fitting to obtain the Weibull model parameters for each known class; the Weibull fitting parameters ρ c Including displacement τ c Shape κ c and scale λ c , used to estimate the probability of abnormal samples during the testing phase; As shown in the formula, the tail length η r This represents the selection of r maximum distances between the training samples and the MAV of each class; r c =(t c ,k c ,l c )=FitHigh(||AV i,c -m c ||,h r ) Next, the Weibull cumulative distribution function, or CDF probability, is prepared for the next process to estimate the class of the test sample; this is done by fitting the parameter ρ. c and distance d i,c Estimate the corresponding CDF i,c It evaluates the relationship between the i-th positive training sample and the c-th class; the formula is shown below: ， The known sample distribution is a monotonically increasing function, and the CDF value increases with increasing distance. Its testing phase: S1: Input the preprocessed test data into the training model, where the activation feature value of the j-th test sample is represented as AV. j ;d j,c This represents the j-th test sample and the c-th known class m. c The distance between them is used to estimate whether the input is "far" from the known training data; d j,c It can be represented as: d j,c =||m c - OFF j ||, S2: To assess the probability that a test sample is an outlier, CDF is used. j,c The value of d is used to determine whether the j-th test sample belongs to the c-th known class; specifically, for the j-th test sample, the distance d j,c and the corresponding fitting parameters ρ c Used to calculate CDF j,c It measures the weights of input samples that do not belong to class c and is used as the core for estimating unknown class rejection; through distance d j,c Distance to the center of each feature class and fitting parameter ρ c The calculated values ​​are concentrated around 1, indicating that it does not belong to any known class. Therefore, when the test sample is of an unknown class, its corresponding CDF value is distributed around 1. Since known classes are included in the test sample, there are K+1 classes in the test sample. The K+1th class represents the unknown class sample, but the CNN model only has the output probabilities of K known classes, i.e., K class nodes, and lacks the K+1th class node. In order to calculate the output probability corresponding to the input sample belonging to the K+1 class nodes, the OpenMax method is used to correct the AV of the input sample, i.e., the AV value corresponding to the K-dimensional class node. Furthermore, the pseudo-AV of the unknown class corresponding to each sample is calculated, i.e., the (K+1)th class node. , ， ， OpenMax method for correcting AV values ​​using AV j The element-wise product of 1 - CDFj is used to calculate the corrected AV; therefore, when CDF j When it approaches 1, the corresponding known class correction for class c is... A value close to 0 indicates that the input test sample is more likely to be of the unknown class, and the expected pseudo-AV of the unknown class is the corresponding The value is greater; However, due to the characteristics of vibration signals, Case Western Reserve University (CRWU) provides very clear data, in which the AV distribution of class nodes for known class samples is relatively strong, all greater than 0; however, due to the lack of prior knowledge about unknown class samples, the AV extraction values ​​of class nodes for unknown class samples show a negative distribution; in this case, the OpenMax method is still used to correct the AV. When the sample is of an unknown class, because the AVj,i values ​​corresponding to the class nodes of that sample are negative, the corresponding CDF... j,i As shown in the above formula, this will result in a pseudo-AV value for the unknown class. It is negative; however, the corrected AV value of the known class is AV. j The value is close to 0, which causes the pseudo AV of the unknown class to be less than the corrected AV of the known class. This is inconsistent with the expectation, so it is easy to misclassify the known class as the unknown class. Therefore, for the bearing fault dataset, the AV value is corrected by using an exponential function, and the pseudo AV value of the unknown class is further calculated to ensure the monotonicity of the extracted AV distribution and make it higher than the positive value. The improved correction formula for AV is expressed as follows: ， ， S3: After obtaining the corrected AV, the probability value corresponding to each class is calculated to determine the class of each sample; however, the unknown class appears in the testing phase of the open set scenario; therefore, it contradicts the probability sum of 1; in order to calculate the probability distribution of each test sample belonging to a known class and the unknown class, the corrected average and the pseudo-average of the known class are used as the numerator, and their sum is used as the denominator, satisfying the requirement of probability sum; therefore, the ability to reject input is provided when the unknown class, i.e., y = K + 1, has the highest probability: 。 2. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent fault diagnosis method for engineering machinery based on the open set hypothesis as described in claim 1.

3. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that executes the intelligent fault diagnosis method for engineering machinery based on the open set hypothesis as described in claim 1.

Images