CN-121980377-A - Method for diagnosing faults of guide bearing of hydroelectric equipment

CN121980377ACN 121980377 ACN121980377 ACN 121980377ACN-121980377-A

Abstract

The invention belongs to the field of machinery, and particularly relates to a fault diagnosis method for a guide bearing of hydroelectric equipment. The method aims at solving the problem that after the new fault type is learned by the fault diagnosis model of the existing hydroelectric equipment guide bearing, the fault diagnosis has high error rate. The hydroelectric equipment guide bearing fault diagnosis method comprises the steps of constructing a hydroelectric equipment guide bearing fault diagnosis pre-training model, obtaining an expansion training set, training the hydroelectric equipment guide bearing fault diagnosis pre-training model according to the expansion training set to obtain a trained hydroelectric equipment guide bearing fault diagnosis model, collecting a hydroelectric equipment vibration sensor signal, and inputting the hydroelectric equipment vibration sensor signal into the trained hydroelectric equipment guide bearing fault diagnosis model to obtain a fault diagnosis result. The problem of the fault diagnosis has the high error rate after the new fault type is learned to the fault diagnosis model of current hydropower equipment guide bearing is solved.

Inventors

JING XIUYAN
WANG ZHENXIN
JI LIANTAO
WANG PU
ZHOU JIAN
TIAN HAIPING
WANG JUN
WANG YANG
REN JISHUN
ZHANG MINWEI

Assignees

国家电网有限公司
北京中元瑞讯科技有限公司
中国电力科学研究院有限公司
国网湖南省电力有限公司
国网湖南省电力有限公司电力科学研究院
国网新源控股有限公司

Dates

Publication Date: 20260505
Application Date: 20251218

Claims (10)

1. The method for diagnosing the faults of the guide bearing of the hydro-electric equipment is characterized by comprising the following steps of: firstly, constructing a hydropower equipment guide bearing fault diagnosis pre-training model; Step two, acquiring an expansion training set, training a pre-training model for diagnosing the guide bearing faults of the hydropower equipment according to the expansion training set, and obtaining a trained model for diagnosing the guide bearing faults of the hydropower equipment; Step three, collecting signals of a vibration sensor of the hydroelectric equipment; And inputting the signals of the vibration sensor of the hydroelectric equipment into a trained failure diagnosis model of the guide bearing of the hydroelectric equipment to obtain a failure diagnosis result.
2. The method for diagnosing the faults of the guide bearing of the hydroelectric equipment according to claim 1 is characterized by comprising the following steps of: Constructing a hydroelectric equipment guide bearing fault diagnosis model; the hydroelectric equipment guide bearing fault diagnosis model comprises a signal coding unit, a text coding unit, a vector fusion unit and a classification unit; step two, a first training set and a second training set are obtained; The first training set comprises X fault vibration signal data samples, wherein X is a positive integer; The second training set comprises Y fault text data samples, wherein Y is a positive integer; step three, taking the first training set and the second training set as the input of a hydroelectric equipment guide bearing fault diagnosis model, and outputting a prediction result by the hydroelectric equipment guide bearing fault diagnosis model; training the hydroelectric equipment guide bearing fault diagnosis model according to the input and output of the hydroelectric equipment guide bearing fault diagnosis model, and stopping training when parameters of the hydroelectric equipment guide bearing fault diagnosis model are converged to obtain a hydroelectric equipment guide bearing fault diagnosis pre-training model.
3. The method for diagnosing a failure of a guide bearing of a hydroelectric installation according to claim 2, wherein in the step one three, the first training set and the second training set are used as inputs of a failure diagnosis model of the guide bearing of the hydroelectric installation, and the failure diagnosis model of the guide bearing of the hydroelectric installation outputs a prediction result, and the specific process is as follows: b1, inputting fault vibration signal data samples in a first training set into a signal coding unit to obtain a signal embedding vector; b2, inputting the fault text data sample in the second training set into a text coding unit to obtain a text embedded vector; b3, inputting the signal embedded vector and the text embedded vector into a vector fusion unit together to obtain a fusion vector; And B4, inputting the fusion vector into a classification unit to obtain a prediction result.
4. The method for diagnosing a failure of a guide bearing of a hydroelectric installation according to claim 3, wherein the step B1 of inputting the failure vibration signal data sample in the first training set into the signal encoding unit to obtain the signal embedding vector comprises the following steps: the signal encoding unit comprises a position encoder and a signal encoder; b1.1, inputting fault vibration signal data into a position encoder for position encoding processing to obtain a position encoded intermediate vector ; B1.2 intermediate vector after position coding The input signal encoder performs signal encoding processing to obtain a signal embedded vector Signal embedding vector The signal output as the signal encoding unit is embedded in the vector.
5. The method for diagnosing a guide bearing failure of a hydroelectric equipment according to claim 4, wherein the data of the failure vibration signal in B1.1 is input into a position encoder for position encoding processing to obtain a position-encoded intermediate vector The specific process is as follows: b1.1.1, the fault vibration signal data X 1 is segmented to obtain segmented signals ; Wherein, the K, m is a positive integer; b1.1.2 to segment signals Inputting the full connection layer to obtain local signal characteristics ; Wherein, the Representing local features of signals The k element of (a); b1.1.3 local features on the signal Inserting class tokens before the first element in a class Obtaining intermediate features ; B1.1.4 intermediate features Performing position coding to obtain a position-coded intermediate vector And the formula is as follows: Wherein, the For embedding the projection matrix, P represents the position-coding matrix.
6. The method for diagnosing a guide bearing failure of a hydro-electric apparatus according to claim 5, wherein the signal encoder in B1.2 includes a first layer encoder, a second layer encoder, a third layer encoder, a fourth layer encoder, a fifth layer encoder, a sixth layer encoder, a seventh layer encoder, an eighth layer encoder, a ninth layer encoder, a tenth layer encoder, an eleventh layer encoder, and a twelfth layer encoder; The first encoder, the second encoder, the third encoder, the fourth encoder, the fifth encoder, the sixth encoder, the seventh encoder, the eighth encoder, the ninth encoder, the tenth encoder, the eleventh encoder and the twelfth encoder are all Transformer encoders; the position-coded intermediate vector The input signal encoder performs signal encoding processing to obtain a signal embedded vector The specific process is as follows: B1.2.1 intermediate vector Sequentially inputting a first layer encoder, a second layer encoder, a third layer encoder, a fourth layer encoder, a fifth layer encoder, a sixth layer encoder, a seventh layer encoder, an eighth layer encoder, a ninth layer encoder, a tenth layer encoder, an eleventh layer encoder and a twelfth layer encoder, wherein the output of the former layer encoder is used as the input of the latter layer encoder, and the output of the first layer encoder is expressed as ;l=[1,2,...12]; The specific process is as follows: Intermediate vector Input to a first layer encoder, the first layer encoder outputs ; Intermediate vector The first layer encoder and the second layer encoder are sequentially input, and the output of the second layer encoder is expressed as ; Intermediate vector The first layer encoder, the second layer encoder and the third layer encoder are sequentially input, and the output of the third layer encoder is expressed as ; Intermediate vector The first layer encoder, the second layer encoder, the third layer encoder and the fourth layer encoder are sequentially input, and the output of the fourth layer encoder is expressed as ; Intermediate vector The first layer encoder, the second layer encoder, the third layer encoder, the fourth layer encoder and the fifth layer encoder are sequentially input, and the output of the fifth layer encoder is expressed as ; Intermediate vector The first layer encoder, the second layer encoder, the third layer encoder, the fourth layer encoder, the fifth layer encoder and the sixth layer encoder are sequentially input, and the output of the sixth layer encoder is expressed as ; Intermediate vector The first layer encoder, the second layer encoder, the third layer encoder, the fourth layer encoder, the fifth layer encoder, the sixth layer encoder and the seventh layer encoder are sequentially input, and the output of the seventh layer encoder is expressed as ; Intermediate vector The first layer encoder, the second layer encoder, the third layer encoder, the fourth layer encoder, the fifth layer encoder, the sixth layer encoder, the seventh layer encoder and the eighth layer encoder are sequentially input, and the output of the eighth layer encoder is expressed as ; Intermediate vector The outputs of the first layer encoder, the second layer encoder, the third layer encoder, the fourth layer encoder, the fifth layer encoder, the sixth layer encoder, the seventh layer encoder, the eighth layer encoder and the ninth layer encoder are sequentially input and expressed as ; Intermediate vector The first layer encoder, the second layer encoder, the third layer encoder, the fourth layer encoder, the fifth layer encoder, the sixth layer encoder, the seventh layer encoder, the eighth layer encoder, the ninth layer encoder and the tenth layer encoder are sequentially input, and the output of the tenth layer encoder is expressed as ; Intermediate vector The first layer encoder, the second layer encoder, the third layer encoder, the fourth layer encoder, the fifth layer encoder, the sixth layer encoder, the seventh layer encoder, the eighth layer encoder, the ninth layer encoder, the tenth layer encoder and the eleventh layer encoder are sequentially input, and the output of the eleventh layer encoder is expressed as ; Intermediate vector The first layer encoder, the second layer encoder, the third layer encoder, the fourth layer encoder, the fifth layer encoder, the sixth layer encoder, the seventh layer encoder, the eighth layer encoder, the ninth layer encoder, the tenth layer encoder, the eleventh layer encoder and the twelfth layer encoder are sequentially input, and the output of the twelfth layer encoder is expressed as ; B1.2.2 output of the first layer encoder Output of the second layer encoder Output of third layer encoder Output of fourth layer encoder Output of fifth layer encoder Output of the sixth layer encoder Output of seventh layer encoder Output of eighth layer encoder Output of the ninth layer encoder Output of tenth layer encoder Output of eleventh layer encoder Output of twelfth layer encoder Inputting the global average pooling layers together to obtain global feature vectors ; B1.2.3 Global feature vector Performing linear change processing to obtain signal embedded vector 。
7. The method for diagnosing a guide bearing failure of a hydroelectric installation according to claim 6, wherein the output of the encoder of the previous layer in B1.2.1 is used as the input of the encoder of the next layer, and the formula is: Wherein, the Representing the self-attention output of the layer I encoder, layerNorm representing layer normalization, conv1D representing one-dimensional convolution; expressed by the formula: wherein MultiHead denotes multi-headed attention.
8. The method for diagnosing a guide bearing failure of a hydroelectric device according to claim 7, wherein the step of inputting the failure text data sample in the second training set into the text encoding unit to obtain the text embedded vector in step B2 comprises the following steps: b2.1, performing word segmentation on the fault text data sample to obtain word segmentation sub-word sequences; b2.2, carrying out word vector processing on the word segmentation processed sub word sequence to obtain sub word vectors; b2.3, carrying out position coding treatment on the sub-word vectors to obtain word vectors after position coding; B2.4, inputting the word vector after position coding into a pre-training language model to obtain a text embedded vector 。
9. The method for diagnosing a guide bearing fault of a hydro-electric apparatus according to claim 8, wherein the vector fusion unit in B3 includes an adapter and a cross-modal fusion layer; the signal embedding vector and the text embedding vector are input into a vector fusion unit together to obtain a fusion vector, and the specific process is as follows: B3.1 embedding signals into vectors Alignment processing is carried out in the input adapter to obtain alignment features And the formula is as follows: In the formula, Representing a dimension-reduction matrix in the adapter, Representing an upbound matrix in the adapter; representing point multiplication, GELU representing an activation function; B3.2 alignment features And text embedding vector Inputting a cross-modal fusion layer, and carrying out fusion processing to obtain a fusion vector, wherein the specific process is as follows: B3.2.1 alignment features Performing layer normalization processing to obtain query vector And the formula is as follows: b3.2.2 computing a query vector Embedding vectors with text The semantic similarity matrix S of (1) is expressed as: Wherein, the Representing a scaling factor; B3.2.3 carrying out Softmax normalization processing on the semantic similarity matrix S to obtain attention weight A; B3.2.4 embedding vectors into text according to the attention weight A Weighting and aggregation processing is carried out to obtain fusion characteristics And the formula is as follows: b3.2.5 based on alignment features And fusion features Obtaining a fusion vector And the formula is as follows: LayerNorm denotes a layer normalization process, Representing the output projection matrix.
10. The method for diagnosing a guide bearing failure of a hydro-electric apparatus according to claim 8, wherein, Acquiring an expansion training set, training a pre-training model for diagnosing the guide bearing faults of the hydroelectric equipment according to the expansion training set, and acquiring a trained model for diagnosing the guide bearing faults of the hydroelectric equipment, wherein the specific process is as follows: the training set expansion comprises Z composite fault samples, wherein Z is a positive integer; S1, freezing all parameters of a signal coding unit, a text coding unit and a classification unit in a guide bearing fault diagnosis pre-training model of hydroelectric equipment; s2, only thawing cross-modal fusion layer parameters of a vector fusion unit in a guide bearing fault diagnosis pre-training model of the hydroelectric equipment; And S3, performing fine tuning training on the cross-modal fusion layer parameters of the vector fusion unit in the pre-training model for diagnosing the guide bearing fault of the hydroelectric equipment by using the composite fault samples in the expansion training set, and stopping training when the cross-modal fusion layer Adapter layer parameters are converged to obtain a trained model for diagnosing the guide bearing fault of the hydroelectric equipment.

Description

Method for diagnosing faults of guide bearing of hydroelectric equipment Technical Field The invention belongs to the field of machinery, and particularly relates to a fault diagnosis method for a guide bearing of hydroelectric equipment. Background About 16 to 20 percent of the global clean energy supply depends on hydroelectric generation, and the stable operation of the large hydroelectric generating set has strategic significance for power grid peak shaving and renewable energy consumption. However, the hydroelectric generating set bears heavy load, variable rotation speed and hydraulic pulsation impact for a long time, and the recessive compound faults (such as fatigue peeling of Babbitt metal layer, looseness of bearing bush and oil film oscillation) of the core supporting component-the water guide bearing are easy to cause malignant accidents. According to statistics, after the vibration of the unit exceeds the standard due to the failure of the water guide bearing, the average power generation efficiency is attenuated by more than 15 percent, the repairing period is longer than 168 hours, and the linkage impact is caused to the frequency adjustment of the power grid. The fault diagnosis of the water guide bearing faces multiple technical bottlenecks, namely, the bearing assembly is deeply placed in a water turbine pit, detection needs to be carried out in a humid closed space (humidity is more than 95%), early damage is difficult to capture by traditional manual inspection, coupling effect of composite faults is obvious, for example, when bearing bush clearance is increased and lubricating oil temperature is abnormally and synergistically acted, a low-frequency whirl component in a vibration signal can cover high-frequency impact characteristics of alloy spalling, so that the misjudgment rate of a single threshold diagnosis model is up to 40%, time from vibration abnormal alarm (T1) to fault accurate positioning (T2) is 60% -70% of the time of unplanned shutdown, a certain large-sized hydroelectric generating set is not timely identified due to spalling of an alloy layer of the thrust guide bearing, diagnosis is delayed for 72 hours, and the generation yield is directly lost. The current mainstream diagnosis method adopts multimode means such as vibration swing degree monitoring, oil metal content analysis, infrared thermal imaging and the like. For example, the improved time-frequency convolution network can identify early stripping of an alloy layer by fusing a bearing bush temperature gradient and a vibration envelope spectrum, and the dynamic graph neural network can early warn oil film whirl instability by combining a water pressure pulsation signal and swing phase data. However, the current research is mainly based on the ideal environment of a laboratory, vibration signal interference of other mutually coupled components often exists in vibration signal data of a guide bearing of a hydroelectric generating set in a real working condition, and the model based on laboratory data driving cannot be focused on effective signals in a targeted manner and can only be matched by using learned data rules. Although the data rule of the composite fault is directly learned by the migratable and generalizable pre-training model so as to perform fault type matching, the full fine tuning training not only needs a large amount of calculation and data quantity, but also can cause disastrous forgetting of the prior learning content after learning a new fault type. Therefore, the fault diagnosis error rate of the guide bearing of the hydro-electric equipment is high. Disclosure of Invention The invention aims to solve the problem that the fault diagnosis has high error rate after the new fault type is learned by the fault diagnosis model of the existing guide bearing of the hydroelectric equipment. The utility model provides a hydroelectric equipment guide bearing fault diagnosis method, which comprises the following steps: firstly, constructing a hydropower equipment guide bearing fault diagnosis pre-training model; The water and electricity equipment guide bearing fault diagnosis pre-training model comprises a pre-training signal encoding unit, a pre-training text encoding unit, a pre-training vector fusion unit and a pre-training classification unit; Step two, acquiring an expansion training set, training a pre-training model for diagnosing the guide bearing faults of the hydropower equipment according to the expansion training set, and obtaining a trained model for diagnosing the guide bearing faults of the hydropower equipment; Step three, collecting signals of a vibration sensor of the hydroelectric equipment; And inputting the signals of the vibration sensor of the hydroelectric equipment into a trained failure diagnosis model of the guide bearing of the hydroelectric equipment to obtain a failure diagnosis result. Further, in the first step, a hydropower equipment guide bearing fault diagnosis pre-traini