Search

CN-121520178-B - Reciprocating compressor fault diagnosis method, device, equipment and storage medium

CN121520178BCN 121520178 BCN121520178 BCN 121520178BCN-121520178-B

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a fault diagnosis method, device, equipment and storage medium for a reciprocating compressor, wherein an unsteady fluctuation base line formed by unsteady fluctuation is fitted by obtaining a relevant parameter of leakage in a cylinder of a target reciprocating compressor and adopting a local weighted regression mode, and extracting pure fault characteristics by decoupling and stripping unsteady fluctuation noise through characteristics, dividing the characteristics according to working condition stages, adapting a load sensing model, strengthening time sequence characteristic capture, and finally dynamically adjusting a classification threshold value through double-stage attention focusing key characteristics and combining working condition compensation to output a diagnosis result. Therefore, the invention realizes the collaborative optimization of the fluctuation noise separation, weak feature extraction and accurate classification of the start-stop transition working condition of the reciprocating compressor based on the characteristic decoupling stripping unsteady state fluctuation noise and the combination of the working condition stage division and the load perception BiGRU, and solves the problem that the traditional diagnosis method cannot adapt to the unsteady state characteristics of the start-stop transition working condition.

Inventors

  • XIAO ZHI

Assignees

  • 成都美迅检测设备有限公司

Dates

Publication Date
20260508
Application Date
20251120

Claims (6)

  1. 1. A fault diagnosis method of a reciprocating compressor, comprising the steps of: acquiring in-cylinder leakage associated parameters of a target reciprocating compressor, constructing an original parameter matrix, and preprocessing the original parameter matrix to obtain a standardized parameter matrix, wherein the method specifically comprises the following steps of: acquiring in-cylinder leakage related parameters of a target reciprocating compressor, wherein the in-cylinder leakage related parameters comprise an air suction pressure sequence, an air discharge pressure sequence and a cylinder wall temperature sequence which are respectively acquired by sensors arranged at an air suction port of the compressor, an air discharge port of the compressor and a cylinder wall; According to the preset sampling frequency, the preset period duration and the preset target start-stop cycle times of the reciprocating compressor, respectively extracting data of a plurality of sampling points in an air suction pressure sequence, an air discharge pressure sequence and a cylinder wall temperature sequence, and constructing an original parameter matrix, wherein the expression is specifically as follows: ; in the formula, Representing the matrix of the original parameters, Is the first The original value of the suction pressure at the moment, Is the first The original value of the exhaust pressure at the moment, Is the first The original value of the cylinder wall temperature at the moment, Indicating the number of sample points to be sampled, ; The method comprises the steps of taking the standardized parameter matrix as input to allocate a local neighborhood window, fitting an unsteady fluctuation base line formed by unsteady fluctuation in a local weighted regression mode, and generating a fault feature matrix by utilizing the standardized parameter matrix and the unsteady fluctuation base line, wherein the method specifically comprises the following steps: based on the load change phase of the target reciprocating compressor in the start-stop cycle, the method comprises the following steps of Dynamically allocating local neighborhood windows, and calculating each moment in the windows by adopting Gaussian kernel functions Constructing a local weight matrix; taking the standardized parameter matrix as input, combining each moment Is used for each moment by adopting a local weighted regression algorithm First, the Parameter fitting unsteady fluctuation baselines to obtain fluctuation baseline fitting values of each parameter; The expression of the fluctuation baseline fitting value of each parameter is specifically as follows: ; in the formula, Is the first Time of day (time) A baseline of unsteady state fluctuation of the individual parameters, Is the first A local weight matrix of the time of day, Is the first The first parameter is at A local data matrix within the time neighborhood window; Will each time instant Is the first of (2) Subtracting a fluctuation baseline fitting value from a standardized parameter value of the parameter, stripping unsteady fluctuation noise, decoupling leakage fault characteristics in the cylinder, and integrating to form a fault characteristic matrix; The expression of decoupling out the leakage fault characteristics in the cylinder is specifically as follows: ; in the formula, Is the first Time of day (time) The post-decoupling fault signature of the individual parameters, Is the first Time of day (time) Normalized parameter values for the individual parameters; wherein the load change stage of the target reciprocating compressor during start-stop cycle comprises initial rising period of 0-3s, rapid rising period of 3-7s, and stable period of 7-10s, and each time Dynamically allocated local neighborhood windows, specifically including local neighborhood windows with 50 data points for initial rise and fast rise periods and local neighborhood windows with 30 data points for stationary periods; According to the fault feature matrix, executing working condition stage self-adaptive division, carrying out alignment, enhancement and splicing treatment on each working condition stage, inputting spliced features into a load perception BiGRU model to obtain a high-dimensional feature matrix output by a load perception BiGRU model, wherein the method specifically comprises the following steps: According to the load change stage of the target reciprocating compressor in the start-stop cycle, dividing the decoupling fault feature matrix into stages to obtain three-stage feature matrices 、 、 For characteristic matrix 、 、 Performing alignment, enhancement and splicing treatment; Acquiring a load normalization value of a target reciprocating compressor in start-stop cycle, calculating a load related door control weight of each moment i through a Sigmoid activation function, and constructing a load sensing BiGRU model of dynamic door control adjustment by utilizing the load related door control weight; The load related door control weight comprises a load related weight of an update door and a load related weight of a reset door, and the expression is specifically as follows: ; ; in the formula, In order to update the load-dependent weights of the gates, To reset the load-dependent weights of the gates, Is the first The normalized value of the load at the moment in time, As a matrix of the load weights, As a result of the bias term, Activating a function for Sigmoid; splice features Inputting a load perception BiGRU model, and combining the load-related door control weight to obtain a high-dimensional feature matrix output by the load perception BiGRU model; and screening and weighting fusion are carried out from the high-dimensional feature matrix by using a stage attention computing mechanism to obtain an optimal feature vector, and the optimal feature vector is input into a multi-label classifier fused with a working condition compensation factor to obtain a fault diagnosis result of the reciprocating compressor.
  2. 2. The fault diagnosis method of reciprocating compressor as claimed in claim 1, wherein the splice feature is formed by a plurality of sub-frames Inputting a load perception BiGRU model, and combining the load-related door control weight to obtain a high-dimensional feature matrix output by the load perception BiGRU model, wherein the method specifically comprises the following steps of: splice features The load perception BiGRU model is input, the load-related gating weight is combined to update the load-perceived updating gate and the resetting gate, and the forward hiding state aiming at the forward GRU and the backward hiding state aiming at the backward GRU in the load perception BiGRU model are updated based on gating output; The expressions of the update gate and the reset gate for the update load sensing are specifically as follows: ; ; The updating expressions of the forward hiding state and the backward hiding state are specifically as follows: ; ; in the formula, The function is activated for Sigmoid, Is the first The decoupling feature vector of the moment in time, Representing the hyperbolic tangent activation function, Is the first The load at the moment is related to the door control weight, The state is hidden for the previous moment, Updating the gate and resetting the gate outputs for load sensing, As a matrix of weights, the weight matrix, As a result of the bias term, For the element-wise multiplication, As a result of the candidate hidden state, The hidden state updated at the current moment; and splicing the forward hidden state and the backward hidden state to form a bidirectional fused high-dimensional feature matrix as the output of a load perception BiGRU model, wherein the expression is specifically as follows: ; in the formula, Is the forward GRU (glass fiber reinforced plastic) A hidden state from time; Is rearward GRU (glass fiber reinforced plastic) A hidden state from time; The hidden state after bidirectional fusion is the hidden state after bidirectional fusion at all moments Constructing a high-dimensional feature matrix 。
  3. 3. The fault diagnosis method of reciprocating compressor as claimed in claim 2, wherein the step of obtaining the fault diagnosis result of the reciprocating compressor comprises the steps of using a phase attention computing mechanism to perform screening and weighted fusion from the high-dimensional feature matrix to obtain an optimal feature vector, inputting the optimal feature vector into a multi-label classifier of a fusion working condition compensation factor, and specifically comprising: High-dimensional characteristic matrix Stage division into Calculating the mean vector of hidden states of each stage Calculating importance weights of each stage through attention scoring function ; Feature matrix of each stage And corresponding stage attention weights Weighted summation is carried out to generate a feature matrix of importance of the fusion stage Weighting feature matrices for stages Calculating the attention score of each moment hidden state, and obtaining moment attention weight through a Softmax function ; Weighting the phase by a feature matrix Hidden state of each moment and corresponding moment attention weight Weighted summation is carried out to generate global optimal feature vector ; According to the load sequence of each stage, calculating the load standard deviation, converting the load standard deviation into a working condition compensation factor, and generating a dynamic adjustment classification threshold value based on the reference threshold value of each leakage grade and the working condition compensation factor; To global optimum feature vector Inputting the multi-label classifier, outputting the prediction probability of each leakage level through the full connection layer and the Sigmoid activation function Using the prediction probability And dynamically adjusting the classification threshold value to generate a fault diagnosis result of the reciprocating compressor.
  4. 4. A fault diagnosis apparatus for a reciprocating compressor, characterized by being adapted to perform the fault diagnosis method for a reciprocating compressor as claimed in any one of claims 1 to 3, comprising: The construction module is used for acquiring the in-cylinder leakage associated parameters of the target reciprocating compressor, constructing an original parameter matrix, and preprocessing the original parameter matrix to obtain a standardized parameter matrix; the generation module is used for taking the standardized parameter matrix as input to allocate a local neighborhood window, fitting an unsteady fluctuation baseline formed by unsteady fluctuation in a local weighted regression mode, and generating a fault feature matrix by utilizing the standardized parameter matrix and the unsteady fluctuation baseline; The dividing module is used for executing working condition stage self-adaptive division according to the fault feature matrix, carrying out alignment, enhancement and splicing treatment on each working condition stage, inputting the spliced features into a load perception BiGRU model, and obtaining a high-dimensional feature matrix output by a load perception BiGRU model; And the diagnosis module is used for screening and weighting fusion from the high-dimensional feature matrix by utilizing a stage attention calculation mechanism to obtain an optimal feature vector, and inputting the optimal feature vector into a multi-label classifier fused with a working condition compensation factor to obtain a fault diagnosis result of the reciprocating compressor.
  5. 5. A reciprocating compressor fault diagnosis apparatus comprising a memory, a processor and a reciprocating compressor fault diagnosis program stored on the memory and operable on the processor, the reciprocating compressor fault diagnosis program when executed by the processor implementing the steps of the reciprocating compressor fault diagnosis method according to any one of claims 1 to 3.
  6. 6. A storage medium having stored thereon a reciprocating compressor fault diagnosis program which when executed by a processor implements the steps of the reciprocating compressor fault diagnosis method of any one of claims 1 to 3.

Description

Reciprocating compressor fault diagnosis method, device, equipment and storage medium Technical Field The present invention relates to the field of fault diagnosis technologies, and in particular, to a fault diagnosis method, apparatus, device, and storage medium for a reciprocating compressor. Background The reciprocating compressor is used as a core power device in industrial scenes such as long-distance pipelines, chemical devices and the like, bears the key tasks of gas compression and transportation, and the running stability of the reciprocating compressor directly influences the safety and the efficiency of the whole production system. In the actual working condition, the compressor needs to frequently respond to process load fluctuation, and is subjected to a start-stop transition working condition, namely the load is rapidly increased from 0 to 30% of rated load within 10 seconds, and in the process, the gas compression in the cylinder is in a strong unsteady state, and core thermodynamic parameters such as suction pressure, exhaust pressure, cylinder wall temperature and the like show severe nonlinear fluctuation. The leakage in the cylinder is the most easily-occurring thermodynamic fault under the working condition, the leakage amount is only 3% -8% of the rated value of early fault, the characteristic signal is easily covered by unsteady fluctuation noise with the amplitude reaching 20% of the rated value, and the variation trends of multiple parameters are mutually coupled, so that the fault characteristics are difficult to identify. The traditional fixed threshold method is dependent on a steady-state working condition parameter range, unsteady fluctuation cannot be adapted, a simple time sequence model (such as a common LSTM) does not consider the stage characteristics of a start-stop working condition, fluctuation noise and fault characteristics are difficult to separate, a method based on machine learning is not optimized for unsteady data, weak coupling characteristics are not recognized, so that the accuracy of in-cylinder leakage diagnosis under the start-stop transition working condition is extremely low, early faults cannot be effectively early-warned, safety accidents such as equipment damage and production interruption are easy to cause, and an accurate diagnosis technical scheme for adapting the start-stop transition working condition characteristics is needed. Disclosure of Invention The invention provides a fault diagnosis method, device and equipment for a reciprocating compressor and a storage medium, and aims to solve at least one technical problem. In order to achieve the above object, the present invention provides a fault diagnosis method for a reciprocating compressor, comprising the steps of: acquiring in-cylinder leakage associated parameters of a target reciprocating compressor, constructing an original parameter matrix, and preprocessing the original parameter matrix to obtain a standardized parameter matrix; using the standardized parameter matrix as input to allocate a local neighborhood window, adopting a local weighted regression mode to fit an unsteady fluctuation baseline formed by unsteady fluctuation, and generating a fault feature matrix by using the standardized parameter matrix and the unsteady fluctuation baseline; according to the fault feature matrix, performing working condition stage self-adaptive division, performing alignment, enhancement and splicing treatment on each working condition stage, inputting spliced features into a load perception BiGRU model, and obtaining a high-dimensional feature matrix output by a load perception BiGRU model; and screening and weighting fusion are carried out from the high-dimensional feature matrix by using a stage attention computing mechanism to obtain an optimal feature vector, and the optimal feature vector is input into a multi-label classifier fused with a working condition compensation factor to obtain a fault diagnosis result of the reciprocating compressor. Optionally, acquiring in-cylinder leakage related parameters of the target reciprocating compressor, and constructing an original parameter matrix, which specifically comprises the following steps: acquiring in-cylinder leakage related parameters of a target reciprocating compressor, wherein the in-cylinder leakage related parameters comprise an air suction pressure sequence, an air discharge pressure sequence and a cylinder wall temperature sequence which are respectively acquired by sensors arranged at an air suction port of the compressor, an air discharge port of the compressor and a cylinder wall; According to the preset sampling frequency, the preset period duration and the preset target start-stop cycle times of the reciprocating compressor, respectively extracting data of a plurality of sampling points in an air suction pressure sequence, an air discharge pressure sequence and a cylinder wall temperature sequence, and constructing an original para