A device fault diagnosis method based on meta learning

Abstract

The application provides a device fault diagnosis method based on meta learning, and belongs to the technical field of device fault diagnosis, and solves the problems that existing device fault knowledge graphs cannot perform fault diagnosis under small sample conditions and that a negative sample generation strategy is unreasonable.A device fault diagnosis method based on meta learning comprises the following steps: extracting time-sequenced device operation signals, performing K-dimensional signal decomposition on the operation signals to obtain signal IMF components, obtaining signal characteristic values according to the signal IMF components, and extracting fault time sequence segments as fault time sequence segment nodes based on the signal characteristic values; performing concept abstraction on device text rule data, defining classification standards according to different fault types, obtaining fault type entity representations based on the classification standards, and taking the fault type entity representations as fault type nodes; and constructing a device fault information knowledge graph based on the fault time sequence segment nodes and the fault type nodes.

Description

Technical Field

[0001] This invention relates to the field of equipment fault diagnosis technology, and in particular to a method for equipment fault diagnosis based on meta-learning. Background Technology

[0002] Due to the complexity of industrial production sites, the accuracy of actual fault diagnosis is far lower than laboratory results. Field-level fault diagnosis requires technologies such as visualization of the same fault cluster and reasoning based on similar faults to assist in decision-making. Traditional data-driven fault diagnosis methods mostly use machine learning or deep learning algorithms to train classification models. While these methods solve the problem of difficulty in building accurate mathematical models due to complex mechanisms, they cannot provide structured fault classification results for each model output, making it difficult to provide structured fault features and classification relationships. In 2012, Google proposed knowledge graph technology. Knowledge graphs can build structured knowledge bases, encapsulating equipment fault information in a knowledge graph format, effectively solving the aforementioned problems. More importantly, knowledge graph reasoning algorithms can solve for unknown facts or relationships in the knowledge graph. By reasoning algorithms to solve for fault causes, knowledge graph-guided equipment fault diagnosis can be achieved. However, accurate reasoning can only be achieved when the causes of equipment faults are clearly recorded. In actual operating conditions, equipment operates under normal conditions for extended periods, and faults are rare, making it difficult to collect and detect fault signals. Acquiring scarce fault signals and integrating them into a fault information knowledge graph requires knowledge graph inference algorithms to perform fault diagnosis tasks with limited sample sizes. To achieve fault inference from a knowledge graph under limited sample conditions, the algorithm needs meta-learning capabilities. However, current technologies do not consider negative sample generation strategies when inferring knowledge entities. While randomly replacing head and tail nodes is commonly used to generate negative samples, in the field of equipment fault diagnosis, randomly generated negative samples are meaningless for training the fault diagnosis model. In fact, using randomly generated negative samples as negative examples of a certain type of fault can cause potentially related faults to lose connection due to learning from negative examples, or the fault diagnosis model may not learn any valuable information from negative examples. For generating negative samples for fault diagnosis, fault entities that are similar to the current triples but should not be classified as positive examples by link prediction algorithms should be generated.

[0003] Therefore, existing equipment fault knowledge graphs suffer from problems such as inability to diagnose faults with small sample sizes and unreasonable negative sample generation strategies. Summary of the Invention

[0004] The purpose of this invention is to provide a device fault diagnosis method based on meta-learning, so as to alleviate the technical problems of existing device fault knowledge graphs being unable to perform fault diagnosis under small sample conditions and having unreasonable negative sample generation strategies.

[0005] In a first aspect, the present invention provides a device fault diagnosis method based on meta-learning, comprising:

[0006] Extract the time-series equipment operation signal, perform K-dimensional signal decomposition on the operation signal to obtain signal IMFs components, obtain signal feature values ​​based on the signal IMFs components, and extract fault time-series segments as fault time-series segment nodes based on the signal feature values;

[0007] The device text rule data is conceptually abstracted, and its classification criteria are defined according to different fault types. Based on the classification criteria, the fault type entity representation is obtained, and the fault type entity representation is used as the fault type node.

[0008] Construct a knowledge graph of equipment fault information based on fault time sequence fragment nodes and fault type nodes;

[0009] Based on the equipment fault information knowledge graph, according to the negative sample generation strategy within the same fault cluster, within a limited range, the negative sample triples of the fault information knowledge graph are queried and generated layer by layer from the starting seed outward.

[0010] The fault link prediction algorithm learns relational metadata of the support set and new instances in the query set from the device fault information knowledge graph through meta-learning.

[0011] The meta-learning-based fault link prediction algorithm determines the fault type and updates the model based on the fault sampling fragments input and the existing fault information in the equipment fault knowledge graph. After obtaining the fault type, it updates the knowledge graph and performs a visual similar fault query in the updated node domain.

[0012] Furthermore, the step of extracting the time-series device operating signal and performing K-dimensional signal decomposition on the operating signal to obtain the signal IMFs components includes:

[0013] Extract timing-sequential device operation signals;

[0014] By finding the optimal solution of the variational model, K-dimensional variational mode decomposition is performed on the time-series device operation signal to obtain the signal IMFs components.

[0015] Furthermore, the step of obtaining the IMFs components of the signal based on the IMFs components includes:

[0016] The signal characteristic values ​​are obtained by calculating the waveform factors of the IMFs components of the signal.

[0017] Furthermore, after the step of conceptually abstracting the device text rule data and defining its classification criteria according to different fault types, the method further includes:

[0018] Extract the classification index sequence Class={C_1,C_2,…,C_n} as the basis for fault classification.

[0019] Furthermore, the equipment fault information knowledge graph is defined as G = (E, R); where E is a set of n entities and R is a set of relationships between entities;

[0020] The equipment fault information knowledge graph includes fault time sequence fragment nodes and fault type nodes;

[0021] The fault timing segment node represents a certain sampling segment in the device timing signal, and the vector representation of the fault timing segment node is the signal feature value;

[0022] The fault type node represents the fault type entity obtained by combining multi-level classification indicators.

[0023] Furthermore, the step of constructing a knowledge graph of equipment fault information based on fault time sequence fragment nodes and fault type nodes includes:

[0024] The definition of a triple in the device fault information knowledge graph constructed based on fault time sequence nodes and fault type nodes is as follows:

[0025] S = {(h,r,t)|h,t∈E,r∈R}

[0026] Where r is the description of the relationship between entity nodes, h is the fault time sequence segment node, and t is the fault type node.

[0027] Furthermore, the step of generating negative sample triples of the fault information knowledge graph based on the equipment fault information knowledge graph, according to the negative sample generation strategy within the same fault cluster, and querying and generating the negative sample triples of the fault information knowledge graph layer by layer from the starting seed outward within a limited range, includes:

[0028] Based on the negative sample generation strategy within the neighborhood of the same fault cluster, the head or tail entity to be replaced is used as a seed in the equipment fault knowledge graph. Other nodes are queried layer by layer outwards from the defined region. The query sets generated by the head node h and tail node t are denoted as ε, respectively. h and ε t Its definition is as follows:

[0029]

[0030]

[0031] in This represents the result of the previous round of diffusion query, i.e., the result ε of the current round of diffusion query. h It is linked from the tail node t of the previous query. Similarly; according to ε h Corresponding node set and ε t Corresponding node set Calculate the similarity margins margin_h and margin_t between each node and the starting node, using the following formulas:

[0032]

[0033]

[0034] The node with the smallest margin_h value is found. The node with the smallest margin_t value The formula for the replacement strategy to obtain negative samples is as follows:

[0035]

[0036] When the head entity is replaced, the formula for the generated negative sample triplet S is as follows:

[0037]

[0038] When the tail entity is replaced, the formula for the generated negative sample triplet S is as follows:

[0039]

[0040] Furthermore, prior to the step of learning the relational metadata of the support set and new instances in the query set of the device fault information knowledge graph through the meta-learning fault link prediction algorithm, the method further includes:

[0041] Based on the knowledge graph of equipment fault information, its triplet dataset is divided into support set training data and query set validation data.

[0042] The support set training data includes multiple task groups, where each task group contains triples of the same relation. Each task group further divides the triples of the same relation into a training support set and a training query set.

[0043] Furthermore, the step of learning the relational metadata of the support set and new instances in the query set of the device fault information knowledge graph through the meta-learning fault link prediction algorithm includes:

[0044] In the support set, for the input triples (h) i ,ti ,r i The relational metadata is calculated using an L-layer neural network. The calculation formula is as follows:

[0045]

[0046] x l =σ(W l x l-1 +b l )

[0047]

[0048] Where L is the number of layers in the neural network and l∈(1,2,…,L-1), W l Let b be the weight of the l-th layer. l This is the bias term for the l-th layer;

[0049] For the extracted relation metadata, the average method is used to calculate the relation metadata for this task. The calculation formula is as follows:

[0050]

[0051] Where K represents the number of relational elements participating in the training for this mission;

[0052] Based on the knowledge graph link prediction algorithm TransE, calculate the tail node of the triple corresponding to the current relation element. The scoring function used here is as follows:

[0053]

[0054] Where ||x|| denotes the calculation of the L-2 norm of vector x, and A lower value indicates a more accurate triplet calculation result, while a higher value indicates that the triplet is a negative sample. For the obtained tail node, the loss function formula is as follows:

[0055]

[0056] in[] + This represents the positive examples of the data, where γ is a hyperparameter representing the distance. The score for negative examples;

[0057] Based on the loss function L(S) r The gradient of this relation element can be calculated to guide parameter changes. (Gradient element) The calculation formula is as follows:

[0058]

[0059] Based on the calculation of gradient elements, relations in the relation query set can be realized, and the new relation element is denoted as... The calculation formula is as follows:

[0060]

[0061] After updating the relation element, the unknown triples (h) can be predicted in the query set. j ,t j ,r j The corresponding fault node of )′, and the scoring function in the query set is as follows:

[0062]

[0063] The loss function is as follows:

[0064]

[0065] In a second aspect, the present invention also provides a computer-readable storage medium storing computer-executable instructions, which, when invoked and executed by a processor, cause the processor to perform the method described in the first aspect.

[0066] This invention provides a device fault diagnosis method based on meta-learning, comprising:

[0067] Extract the time-series equipment operation signal, perform K-dimensional signal decomposition on the operation signal to obtain signal IMFs components, obtain signal feature values ​​based on the signal IMFs components, and extract fault time-series segments as fault time-series segment nodes based on the signal feature values;

[0068] The device text rule data is conceptually abstracted, and its classification criteria are defined according to different fault types. Based on the classification criteria, the fault type entity representation is obtained, and the fault type entity representation is used as the fault type node.

[0069] Construct a knowledge graph of equipment fault information based on fault time sequence fragment nodes and fault type nodes;

[0070] Based on the equipment fault information knowledge graph, according to the negative sample generation strategy within the same fault cluster, within a limited range, the negative sample triples of the fault information knowledge graph are queried and generated layer by layer from the starting seed outward.

[0071] The fault link prediction algorithm learns relational metadata of the support set and new instances in the query set from the device fault information knowledge graph through meta-learning.

[0072] The meta-learning-based fault link prediction algorithm determines the fault type and updates the model based on the fault sampling fragments input and the existing fault information in the equipment fault knowledge graph. After obtaining the fault type, it updates the knowledge graph and performs a visual similar fault query in the updated node domain.

[0073] The device fault diagnosis method based on meta-learning provided by this invention first extracts signal feature information from device operation data, then extracts fault rule chains from the text description of the device data into a knowledge graph. By fusing device signal features and text rules, a device fault information knowledge graph is constructed. A meta-learning fault link prediction algorithm is then used to learn this knowledge graph. This enables the learning of relational metadata based on the relationships between fault chain triples, even with small sample sizes. Link prediction is then performed on both the support set and the query set, outputting inference results for the fault type. This solves the problem of scarce fault data in actual operating conditions, making fault diagnosis impossible with limited sample sizes. Furthermore, this scheme employs a negative sample generation strategy within the same fault cluster neighborhood, generating similar fault entities as replacements for negative samples. This avoids the problem of random replacement of head and tail entities, which prevents the prediction algorithm from learning useful knowledge from negative examples. This effectively solves the problem of unreasonable negative sample generation strategies and further improves the accuracy of fault diagnosis.

[0074] Correspondingly, the computer-readable storage medium provided by the present invention also has the above-mentioned technical effects. Attached Figure Description

[0075] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0076] Figure 1 A flowchart of a device fault diagnosis method based on meta-learning provided in an embodiment of the present invention;

[0077] Figure 2 This is a schematic diagram of a device fault diagnosis method based on meta-learning in an embodiment of the present invention;

[0078] Figure 3 This is a schematic diagram illustrating the signal decomposition and solution of signal components and the calculation of signal features in an embodiment of the present invention;

[0079] Figure 4 This is a schematic diagram of the equipment fault information knowledge graph structure in an embodiment of the present invention;

[0080] Figure 5This is a schematic diagram of the negative sample generation strategy diffusion query method in an embodiment of the present invention;

[0081] Figure 6 This is a schematic diagram of the triplet partitioning method for the meta-learning training task in an embodiment of the present invention;

[0082] Figure 7 This is a schematic diagram of the MetaF fault link prediction algorithm based on meta-learning in an embodiment of the present invention;

[0083] Figure 8 This is a schematic diagram of the fault diagnosis and similar fault list query process of MetaF in an embodiment of the present invention;

[0084] Figure 9 This is a schematic diagram of similar fault query results in an embodiment of the present invention;

[0085] Figure 10 This is a schematic diagram of similar fault scoring results in an embodiment of the present invention. Detailed Implementation

[0086] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] The terms "comprising" and "having," and any variations thereof, used in the embodiments of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or devices.

[0088] Existing equipment fault knowledge graphs suffer from problems such as inability to diagnose faults with small sample sizes and unreasonable negative sample generation strategies.

[0089] To solve the above problems, such as Figure 1 and Figure 2 As shown, this embodiment of the invention provides a device fault diagnosis method based on meta-learning, including:

[0090] S1: Extract the time-series equipment operation signal, perform K-dimensional signal decomposition on the operation signal to obtain signal IMFs components, obtain signal feature values ​​based on the signal IMFs components, and extract fault time-series segments as fault time-series segment nodes based on the signal feature values;

[0091] S2: Abstract the device text rule data conceptually, define classification criteria according to different fault types, obtain fault type entity representations based on classification criteria, and use fault type entity representations as fault type nodes.

[0092] S3: Construct a knowledge graph of equipment fault information based on fault time sequence fragment nodes and fault type nodes;

[0093] S4: Based on the equipment fault information knowledge graph, according to the negative sample generation strategy within the same fault cluster, within a limited range, query and generate negative sample triples of the fault information knowledge graph layer by layer from the starting seed.

[0094] S5: Learn the relational metadata of the support set and new instances in the query set of the device fault information knowledge graph through the meta-learning fault link prediction algorithm;

[0095] S6: The meta-learning-based fault link prediction algorithm determines the fault type and updates the model based on the fault sampling fragments input and the existing fault information in the equipment fault knowledge graph. After obtaining the fault type, it updates the knowledge graph and performs a visual similar fault query in the updated node domain.

[0096] The device fault diagnosis method based on meta-learning provided in this invention first extracts signal feature information from device operation data, then extracts fault rule chains from the text description of the device data into a knowledge graph. By fusing device signal features and text rules, a device fault information knowledge graph is constructed. A meta-learning fault link prediction algorithm is then used to learn this knowledge graph. This enables the learning of relational metadata based on the relationships between fault chain triples even with small sample sizes, and achieves link prediction in both the support set and query set, outputting inference results for fault types. This solves the problem of scarce fault data in actual operating conditions, making fault diagnosis impossible with limited sample sizes. Furthermore, this scheme employs a negative sample generation strategy within the same fault cluster neighborhood, generating similar fault entities as replacements for negative samples. This avoids the problem of random replacement of head and tail entities, which prevents the prediction algorithm from learning useful knowledge from negative examples, effectively solving the problem of unreasonable negative sample generation strategies and further improving the accuracy of fault diagnosis.

[0097] In one possible implementation, step S1 specifically includes:

[0098] Extract timing-sequential device operation signals;

[0099] By finding the optimal solution of the variational model, K-dimensional variational mode decomposition is performed on the time-series device operation signal to obtain the signal IMFs components.

[0100] Extracting time-series device operating signals and performing K-dimensional variational mode decomposition on them, this method determines the IMFs components of the signal by finding the optimal solution of the variational model. The process is as follows: Figure 3 As shown, the specific steps are as follows:

[0101] 1-1) Solve the constrained variational problem, the formula is as follows:

[0102]

[0103] Where {u k} represents the IMFs components obtained from the decomposition, {ω k} represents the center frequency of the IMFs component, and K represents the number of IMFs decomposed.

[0104] 1-2) The constrained problem is transformed into an unconstrained problem by using a quadratic penalty factor β and a Lagrange multiplier λ, as shown in the following formula:

[0105]

[0106] 1-3) Solve the problem in 1-2) using the alternating direction multiplier method, as shown in the following formula:

[0107]

[0108] In the above formula, n is the number of iterations.

[0109] 1-4) Iterate 1-3) repeatedly until the convergence condition is met. The convergence condition formula is as follows:

[0110]

[0111] 1-5) Finally, the sequence f is decomposed into K signal components, and the corresponding sequence is denoted as {IMFs1, IMFs2, ..., IMFs}. k}

[0112] To obtain time-series signal feature data from the knowledge graph based on K signal components, it is necessary to solve for the waveform factor of each IMF component to obtain its one-dimensional signal feature value. The waveform factor is the ratio of the root mean square value of the signal to the rectified average value. The calculation formula is as follows:

[0113]

[0114] Based on K modal components {IMFs1, IMFs2, ..., IMFs} k The waveform factor matrix {form1,form2,…,form} can be calculated. k The formula for calculating the length L of the k-th dimension waveform factor vector is as follows:

[0115]

[0116] Where length(f) represents the length of the signal sequence f, v represents the frequency, and n represents the rotational speed.

[0117] In one possible implementation, steps S2 and S3 specifically include:

[0118] The device text data is conceptually abstracted, classification criteria are defined for different fault types, and a knowledge graph of device fault information is constructed. The index sequence Class={C1,C2,…,C…} used as the classification basis is extracted. n The equipment fault information knowledge graph is defined as G = (E, R), where E is a set of n entities and R is a set of relationships between entities. Two types of nodes are defined in the equipment fault information knowledge graph: fault time sequence nodes and fault type nodes. A fault time sequence node represents a sampling segment in the equipment's time-series signal; its vector representation is the feature vector obtained after mode decomposition. A fault type node represents the entity representation of the fault type obtained by combining multi-level fault classification indicators. The definitions of triples in the equipment fault information knowledge graph obtained from the classification indicators and graph nodes are as follows:

[0119] S = {(h,r,t)|h,t∈E,r∈R}

[0120] Where r represents the relationship description between entity nodes, h represents the fault time sequence node, and t represents the fault type node. Fault feature values ​​and fault rule chains are combined, and the correspondence between graph nodes is <fault time sequence node, r, fault type node>. This completes the construction of the equipment fault information knowledge graph, whose structure is as follows: Figure 4 As shown, it consists of two parts: the fault map rule layer and the fault map data layer, which store the fault rule chain and fault feature value, respectively.

[0121] In one possible implementation, step S4 specifically includes:

[0122] Based on the negative sample generation strategy within the neighborhood of the same fault cluster, the propagation mode of water ripples was simulated. In the knowledge graph, the head or tail entity to be replaced is used as a seed. Within a defined area, the similarity between other nodes and the node to be replaced is queried layer by layer outward from the seed. The diffusion query method is as follows: Figure 5 As shown.

[0123] The strategy for generating negative samples uses the head and tail nodes of the triple (h,r,t) to be replaced as query seeds, and sequentially queries adjacent graph nodes. The query sets generated by the head node h and tail node t are denoted as ε. h and ε t Its definition is as follows:

[0124]

[0125]

[0126] in This represents the result of the previous round of diffusion query, i.e., the result ε of the current round of diffusion query. h It is linked from the tail node t of the previous query. Similarly, then based on ε... h Corresponding node set and ε t Corresponding node set Calculate the similarity between each node and the starting node. Based on the cosine similarity of the embedding vectors in the plane, obtain the similarities between the head node and the query set nodes as margin_h and margin_t, respectively. The calculation formulas are as follows:

[0127]

[0128]

[0129] The node with the smallest margin_h value is found. The node with the smallest margin_t value The formula for the replacement strategy to obtain negative samples is as follows:

[0130]

[0131] When the head entity is replaced, the generated negative sample triple S is calculated as follows:

[0132]

[0133] When replacing the tail entity, the generated negative sample triple S is calculated as follows:

[0134]

[0135] In one possible implementation, prior to the step of learning the relational metadata of the support set and new instances in the query set of the device fault information knowledge graph through the meta-learning fault link prediction algorithm, the method further includes:

[0136] Based on the knowledge graph of equipment fault information, its triplet dataset is divided into support set training data and query set validation data.

[0137] The support set training data includes multiple task groups, where each task group contains triples of the same relation. Each task group further divides the triples of the same relation into a training support set and a training query set.

[0138] Based on the data format of the equipment fault information knowledge graph, its triples need to be partitioned into meta-learning tasks. The method for partitioning triples for meta-learning training tasks is as follows: Figure 6 As shown. For the triplet dataset D, it is divided into a support set and a training data set D. train And query set validation data D test and D train Further divided into n task groups, namely Task1, Task2, ..., Task n For each task group, triplets of the same relation are selected within it, and these are used as the training support set S for that task group. train and training query set Q train .

[0139] In one possible implementation, step S5 specifically includes:

[0140] MetaF, a fault link prediction algorithm based on meta-learning, is driven by information from a knowledge graph of equipment fault information. It learns relational meta-information of the support set in a relational meta-learner and performs task learning of new instances in the query set. It diagnoses equipment faults of unknown fault types even with small sample sizes. Its algorithm structure is as follows: Figure 7 As shown, the specific steps are as follows:

[0141] 2-1) In the support set, for the input triples (h) i ,t i ,r i The relational metadata is calculated using an L-layer neural network. The calculation formula is as follows:

[0142]

[0143] x l =σ(W l x l-1 +b l )

[0144]

[0145] Where L is the number of layers in the neural network and l∈(1,2,…,L-1), W l Let b be the weight of the l-th layer. l This is the bias term for the l-th layer;

[0146] 2-2) For the extracted relation metadata, the average method is used to calculate the relation metadata for this task. The calculation formula is as follows:

[0147]

[0148] Where K represents the number of relational elements participating in the training for this mission;

[0149] 2-3) Based on the knowledge graph link prediction algorithm TransE, calculate the tail node of the triple corresponding to the current relation element. The scoring function used here is as follows:

[0150]

[0151] Where ||x|| denotes the calculation of the L-2 norm of vector x, and A lower value indicates a more accurate triplet calculation result, while a higher value indicates that the triplet is a negative sample. For the obtained tail node, the loss function formula is as follows:

[0152]

[0153] in[] + This represents the positive examples of the data, where γ is a hyperparameter representing the distance. The score for negative examples;

[0154] 2-4) Based on the loss function L(S) r The gradient of this relation element can be calculated to guide parameter changes. (Gradient element) The calculation formula is as follows:

[0155]

[0156] Based on the calculation of gradient elements, relations in the relation query set can be realized, and the new relation element is denoted as... The calculation formula is as follows:

[0157]

[0158] 2-5) After updating the relation element, the unknown triples (h) can be predicted in the query set. j ,t j ,r j The corresponding fault node of )′, and the scoring function in the query set is as follows:

[0159]

[0160] The loss function is as follows:

[0161]

[0162] This invention also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are invoked and executed by a processor, the computer-executable instructions cause the processor to perform the method provided in the above embodiments.

[0163] This invention also provides experimental verification of a device fault diagnosis method based on meta-learning. The specific experimental content is as follows:

[0164] 1. Data Description

[0165] The experimental data is from the rolling bearing failure dataset of Case Western Reserve University (CWRU). The data includes vibration signal data collected at the drive, fan, and base ends of the bearing test bench. Single-point damage was achieved using electrical discharge machining (EDM). The damage diameters were 0.007 inches, 0.014 inches, and 0.021 inches, and the damage locations included the inner ring, outer ring, and rolling elements. The outer ring damage points were located at the 3 o'clock, 6 o'clock, and 12 o'clock positions.

[0166] Experiment 1: Construction of a Knowledge Graph for Equipment Fault Information

[0167] The experiment selected all 12kHz data points from the CWRU dataset. Based on the textual rule description of the dataset, classification was performed, and five categories were summarized to divide the data in the dataset: fault endpoint, fault location, damage depth, equipment operating load, and fault monitoring location. The classification rules for the dataset are described based on these five categories, as shown in Table 1.

[0168] Table 1. CWRU Dataset Classification Rules

[0169]

[0170]

[0171] Based on the above classification rules, the correspondence between nodes in the graph can be obtained. Based on the attribute values ​​of each fault sampling segment in five dimensions, the connection relationship between fault time sequence segment nodes and fault type nodes can be obtained, as shown in Table 2:

[0172] Table 2. Correspondence between nodes

[0173]

[0174] After obtaining the node definitions and data relationships of the equipment fault information knowledge graph, the bearing fault signal needs to be segmented and subjected to variational mode decomposition. The generated feature vectors are combined with the nodes to construct the knowledge graph. The obtained knowledge graph data is stored in a txt file, and the rule-layer fault rule chain in the equipment fault information knowledge graph is stored in the Neo4j database. Based on the above method, after random uniform sampling of the CWRU dataset, 1196 entities and 6 types of relations were extracted. This data stores the association methods of CWRU bearing fault data in five dimensions: endpoint, fault location, damage depth, operating load, and detection location, and realizes knowledge links between fault segments under different types of faults. The distance between any two fault time-series segment nodes in the knowledge graph is two hops, connected by an associated fault type node. This structure facilitates linking and predicting the true fault type of fault segments within a certain region.

[0175] Experiment 2 MetaF Fault Diagnosis and Recommended List of Similar Faults

[0176] MetaF implements the following process for fault diagnosis and querying similar faults: Figure 8 As shown, after inputting a fault sampling fragment, MetaF determines the sample fault type and updates the model based on existing fault information in the knowledge graph. After obtaining the fault type, it updates the knowledge graph and finally performs a visual query for similar faults within the updated node's domain. The experiment uses Python 3.7 and PyTorch 1.7.1 packages for model training, with an NVIDIA GeForce GTX 1060 GPU and Neo4j version community-3.3.1. The experiment randomly selects a sampling fragment from the CWRU validation dataset as input, with a sampling length of 1600. After modal decomposition of this sampling fragment, it is input into the MetaF model for fault type diagnosis, and a query for a list of similar faults is performed based on the diagnosis results. The query results are shown below. Figure 9 As shown. Based on the input fault fragment, a target similar fault cluster and three other similar fault clusters can be retrieved.

[0177] For the retrieved list of similar faults, the similarity of the faults is scored based on the Euclidean distance between the fault vectors. For any set of fault vectors P = {p1, p2, ..., p...} k} and Q = {q1,q2,…,q k The formula for calculating the fault similarity (Fault_Similarity) is as follows:

[0178]

[0179] Based on the input fault segment, calculate the fault similarity with each item in the obtained list of similar faults, and obtain the scoring results as follows: Figure 10 As shown in the table. According to the scoring results, 4 sets of data are most similar to the input fault. Further table lookup shows that these 4 sets of data and the input fault are all of the fault type faultType={point:Drive,located:B,deepth:007,load:1}. Among the 14 similar faults in the inference, 4 fault samples are found to be similar to the input fault, so the sample detection rate of the target similar fault cluster is said to be 28.6%. To further illustrate that MetaF's fault reasoning ability is significantly improved compared to MetaR, the sample detection rate of the target similar fault cluster in the similar fault list reasoning of the two models is compared. The experiments were set up with similar fault list lengths of 10, 15, and 20 respectively. Each experiment was conducted 10 times and the average of the results was taken. The experimental results are shown in Table 3:

[0180] Table 3. Sample detection rate (%) of similar fault clusters.

[0181]

[0182] The experimental results in Table 3 show that MetaF has significantly improved its ability to diagnose faults and detect similar faults. This indicates that the addition of the strategy of generating negative samples in the neighborhood of the same fault cluster effectively improves the model's fault diagnosis capability and effectively assists in fault decision-making through similar faults.

[0183] in conclusion

[0184] To address the challenge of limited fault samples and difficulty in constructing fault information knowledge graphs under real-world operating conditions, this paper proposes a meta-learning-based method for constructing and reasoning about equipment fault knowledge graphs. This method comprises two parts. The first part extracts classification rules from the fault data text descriptions, constructs a fault rule chain for the ontology model, and extracts feature information from the fault data as data information for the fault rule chain, thus constructing the equipment fault information knowledge graph. The second part utilizes a negative sample generation strategy within the neighborhood of the same fault cluster, employing the MetaF model for fault diagnosis and similar fault list reasoning under small sample conditions, thereby assisting in fault diagnosis decision-making based on equipment operating data. Experimental results show that MetaF effectively improves the accuracy of reasoning on the knowledge graph compared to the previous method and can effectively utilize equipment operating data to construct a fault diagnosis domain knowledge graph, demonstrating that the MetaF algorithm can effectively achieve equipment fault diagnosis and similar fault reasoning.

[0185] The apparatus provided in this embodiment of the invention can be specific hardware on a device or software or firmware installed on the device. The implementation principle and technical effects of the apparatus provided in this embodiment of the invention are the same as those in the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the apparatus embodiments can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, apparatuses, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0186] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0187] For example, the division of units is merely a logical functional division; in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be indirect couplings or communication connections through some communication interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0188] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0189] In addition, the functional units in the embodiments provided by the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0190] If the aforementioned functions are implemented as 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 this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0191] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0192] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention. All should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A device fault diagnosis method based on meta-learning, characterized in that, include: Extract the time-series equipment operation signal, perform K-dimensional signal decomposition on the operation signal to obtain signal IMFs components, obtain signal feature values ​​based on the signal IMFs components, and extract fault time-series segments as fault time-series segment nodes based on the signal feature values; The device text rule data is conceptually abstracted, and its classification criteria are defined according to different fault types. Based on the classification criteria, the fault type entity representation is obtained, and the fault type entity representation is used as the fault type node. Construct a knowledge graph of equipment fault information based on fault time sequence fragment nodes and fault type nodes; Based on the equipment fault information knowledge graph, according to the negative sample generation strategy within the same fault cluster, within a limited range, the negative sample triples of the fault information knowledge graph are queried and generated layer by layer from the starting seed outward. The fault link prediction algorithm learns relational metadata of the support set and new instances in the query set from the device fault information knowledge graph through meta-learning. The meta-learning-based fault link prediction algorithm determines the fault type and updates the model based on the fault sampling fragments input and the existing fault information in the equipment fault knowledge graph. After obtaining the fault type, it updates the knowledge graph and performs a visual similar fault query in the updated node domain.

2. The device fault diagnosis method based on meta-learning according to claim 1, characterized in that, The step of extracting the time-series equipment operating signal and performing K-dimensional signal decomposition on the operating signal to obtain the signal IMFs components includes: Extract timing-sequential device operation signals; By finding the optimal solution of the variational model, K-dimensional variational mode decomposition is performed on the time-series device operation signal to obtain the signal IMFs components.

3. The device fault diagnosis method based on meta-learning according to claim 1, characterized in that, The step of obtaining the IMFs components of a signal based on the IMFs components of the signal includes: The signal characteristic values ​​are obtained by calculating the waveform factors of the IMFs components of the signal.

4. The device fault diagnosis method based on meta-learning according to claim 1, characterized in that, After the step of conceptually abstracting the device text rule data and defining its classification criteria according to different fault types, the method further includes: Extract the classification index sequence as the basis for fault classification. .

5. The device fault diagnosis method based on meta-learning according to claim 4, characterized in that, The equipment fault information knowledge graph is defined as G = (E, R); where E is a set of n entities and R is a set of relationships between entities. The equipment fault information knowledge graph includes fault time sequence fragment nodes and fault type nodes; The fault timing segment node represents a certain sampling segment in the device timing signal, and the vector representation of the fault timing segment node is the signal feature value; The fault type node represents the fault type entity obtained by combining multi-level classification indicators.

6. The device fault diagnosis method based on meta-learning according to claim 5, characterized in that, The steps for constructing a knowledge graph of equipment fault information based on fault time sequence fragment nodes and fault type nodes include: The definition of a triple in the device fault information knowledge graph constructed based on fault time sequence nodes and fault type nodes is as follows: ; Where r is the description of the relationship between entity nodes, h is the fault time sequence segment node, and t is the fault type node.

7. The device fault diagnosis method based on meta-learning according to claim 1, characterized in that, The steps of generating negative sample triples for the fault information knowledge graph based on the equipment fault information knowledge graph, according to the negative sample generation strategy within the same fault cluster, and querying and generating the negative sample triples of the fault information knowledge graph layer by layer from the starting seed outward within a limited area, include: Based on the negative sample generation strategy within the neighborhood of the same fault cluster, the head or tail entity to be replaced is used as a seed in the equipment fault knowledge graph. Other nodes are queried layer by layer outwards from the defined region. The query sets generated by the head node h and tail node t are denoted as follows: and Its definition is as follows: ; in This represents the result of the previous round of diffusion query, which is the result of the current round of diffusion query. It is linked from the tail node t of the previous query. Similarly; according to Corresponding node set and Corresponding node set Calculate the similarity between each node and the starting node. and The calculation formula is as follows: ; According to the obtained The node with the smallest value and The node with the smallest value The formula for the replacement strategy to obtain negative samples is as follows: ; When the head entity is replaced, the formula for the generated negative sample triplet S is as follows: ; When the tail entity is replaced, the formula for the generated negative sample triplet S is as follows: .

8. The device fault diagnosis method based on meta-learning according to claim 1, characterized in that, The fault link prediction algorithm, which learns from the knowledge graph of device fault information through meta-learning, supports... Before the steps of holding the relational metadata of the set and creating new instances in the query set, the following are also included: Based on the knowledge graph of equipment fault information, its triplet dataset is divided into support set training data and query set validation data. The support set training data includes multiple task groups, where each task group contains triples of the same relation. Each task group further divides the triples of the same relation into a training support set and a training query set.

9. The device fault diagnosis method based on meta-learning according to claim 1, characterized in that, The steps of learning relational metadata of the support set and new instances in the query set of the fault link prediction algorithm through meta-learning include: in the support set, for the input triples The relational metadata is calculated through an L-layer neural network. The calculation formula is as follows: ; Where L is the number of neural network layers and , For the weights of the l-th layer, For the l-th layer bias term; For the extracted multiple relation meta-information The formula for calculating the relational metadata for this task using the averaging method is as follows: ; Where K represents the number of relational elements participating in the training for this mission; Based on the knowledge graph link prediction algorithm TransE, calculate the tail node of the triple corresponding to the current relation element. The scoring function used here is as follows: ; in This indicates the computation of vector x. The norm is used, and the lower the value, the more accurate the triplet calculation result; conversely, the higher the value, the more accurate the triplet is. For the obtained tail nodes, the loss function formula is as follows: ; in This represents the positive examples of the data. For distance hyperparameters, Scoring for negative examples; Based on the calculation of the loss function L(Sr), the gradient of this relation element is calculated to guide parameter changes. The formula for calculating the gradient element is as follows: ; Based on the calculation of gradient elements, the relations in the relation query set are realized, and the new relation element is denoted as... The calculation formula is as follows: ; After updating the relation element, the unknown triples can be predicted in the query set. The corresponding faulty node, and the scoring function in the query set is as follows: ; The loss function is as follows: .

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 9.

Images