Intelligent fault diagnosis method for reciprocating compressor indicator diagram based on multi-feature fusion

Abstract

This invention relates to the field of intelligent fault diagnosis technology, and provides an intelligent fault diagnosis method for reciprocating compressors based on multi-feature fusion, comprising: acquiring dynamometer card data during the operation of the reciprocating compressor and preprocessing it; extracting multi-dimensional features with clear physical meaning from the dynamometer card, including energy, dynamic, geometric, loss, and proportional features; fusing the multi-dimensional features based on an adaptive weight allocation strategy to calculate a comprehensive fault score; combining feature combinations and weight configurations to match a fault mode library and identify the fault type; and determining the fault level based on the comprehensive fault score using fuzzy boundary processing and a dynamic threshold adaptive update mechanism. This method, by constructing a multi-dimensional feature space and integrating physical mechanisms and intelligent algorithms, achieves high-precision, multi-type, and refined fault diagnosis for reciprocating compressors, significantly improving diagnostic accuracy and reducing the risk of false alarms and missed alarms.

Description

Technical Field

[0001] This invention relates to the field of intelligent fault diagnosis technology, and more specifically, to an intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion. Background Technology

[0002] Reciprocating compressors are core power equipment in industries such as petrochemicals, natural gas processing, and refrigeration and air conditioning. Their operational reliability directly affects the safety and economic efficiency of the entire production system. Statistics show that the downtime cost of a reciprocating compressor can reach tens of thousands to hundreds of thousands of yuan per hour. Therefore, establishing an efficient and accurate fault diagnosis system has significant engineering value and economic importance.

[0003] Dynamometer diagrams (PV diagrams), as important tools reflecting the working cycle characteristics of reciprocating compressors, can intuitively display the internal thermodynamic processes of the compressor, including the four basic stages of intake, compression, exhaust, and expansion. By analyzing parameters such as the shape, area, and key point locations of the PV diagram, the operating status and potential faults of the equipment can be effectively identified. However, traditional PV analysis methods mainly rely on manual interpretation by experienced technicians, which has the following significant drawbacks:

[0004] 1. Highly subjective and lacking consistency

[0005] Manual interpretation of results largely depends on the operator's experience and knowledge background. Different personnel may arrive at different diagnostic conclusions from the same indicator diagram, lacking objectivity and consistency. Especially in the early stages of a fault, subtle changes in characteristics are often difficult to detect with the naked eye, easily leading to missed or misdiagnosed cases.

[0006] 2. Single feature extraction dimension

[0007] Existing diagnostic methods typically focus only on one or a few characteristic parameters of the indicator diagram, such as area and peak pressure, which cannot comprehensively reflect the complex operating state of the compressor. In reality, different types of faults often manifest as a combination of changes in multiple characteristic parameters, and a single feature is insufficient to accurately distinguish the fault type and severity.

[0008] 3. Lack of intelligent weight allocation mechanism

[0009] In multi-feature fusion diagnostics, the sensitivity and importance of different feature parameters vary significantly for different fault types. Traditional methods typically employ fixed weights or simple linear combinations, which cannot dynamically adjust the weight allocation based on actual operating conditions and fault characteristics, thus limiting diagnostic accuracy.

[0010] 4. The fault severity classification is coarse.

[0011] Existing methods typically only provide a binary judgment of "normal" or "abnormal," or use a simple three-level classification (minor, moderate, severe), which cannot meet the needs of modern industry for refined maintenance management. The lack of quantitative assessment of fault severity makes it difficult to formulate reasonable maintenance strategies and timing.

[0012] 5. Insufficient ability to identify multiple fault types

[0013] Common fault types in reciprocating compressors include excessive cylinder clearance, malfunctioning intake and exhaust valves, worn piston rings, and blocked pipes, among others. Each fault exhibits different characteristics on the indicator diagram. Existing methods often only identify one or a few fault types, lacking a unified framework for multi-fault identification.

[0014] 6. Insufficient real-time performance and adaptability

[0015] Industrial environments require fault diagnosis systems with real-time response capabilities, enabling continuous monitoring and timely detection of anomalies during equipment operation. Furthermore, different compressor models and operating conditions exhibit varying operating characteristics, necessitating robust self-adaptability from the diagnostic system. Existing methods fall short in both of these aspects.

[0016] In recent years, with the rapid development of technologies such as artificial intelligence and machine learning, data-driven intelligent fault diagnosis methods have gradually emerged. However, most existing intelligent diagnosis methods directly borrow general algorithms from other fields and lack specific design for the dynamometer card characteristics of reciprocating compressors. Improvements are still needed in terms of the physical meaning of feature extraction, the rationality of weight allocation, and the accuracy of fault classification.

[0017] Therefore, there is an urgent need to develop an intelligent fault diagnosis solution specifically for the indicator diagram of reciprocating compressors. Summary of the Invention

[0018] This invention addresses the technical problems existing in the prior art by providing an intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion. It has core functions such as multi-dimensional feature fusion, adaptive weight allocation, accurate level determination, and multi-fault type identification, which can accurately identify multiple fault types and achieve accurate fault level determination, thereby improving the accuracy and reliability of diagnosis.

[0019] According to a first aspect of the present invention, a method for intelligent fault diagnosis of reciprocating compressor indicator diagrams based on multi-feature fusion is provided, comprising:

[0020] Extracting multidimensional features from the indicator diagram of a reciprocating compressor;

[0021] The multidimensional features are fused based on an adaptive weight allocation strategy, and a comprehensive fault score is calculated.

[0022] The fault type is diagnosed based on the feature combination and weight configuration in the fused features;

[0023] Based on the comprehensive fault score, the fault level is determined by combining fuzzy boundary processing and dynamic threshold adaptive update mechanism.

[0024] Based on the above technical solution, the present invention can also be improved as follows.

[0025] Optionally, the process may also include preprocessing the indicator diagram.

[0026] According to a second aspect of the present invention, a reciprocating compressor indicator diagram intelligent fault diagnosis system based on multi-feature fusion is provided, comprising:

[0027] The acquisition module is used to extract multi-dimensional features from the indicator diagram of the reciprocating compressor;

[0028] The weight allocation and scoring module is used to fuse the multi-dimensional features based on an adaptive weight allocation strategy and calculate a comprehensive fault score.

[0029] The fault type diagnosis module is used to diagnose fault types based on the feature combinations and weights in the fused features.

[0030] The fault classification module is used to determine the fault level based on the comprehensive fault score, combined with fuzzy boundary processing and dynamic threshold adaptive update mechanism.

[0031] According to a third aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the processor is configured to implement the steps of the above-described intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion when executing a computer management program stored in the memory.

[0032] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer management program is stored, wherein when the computer management program is executed by a processor, the steps of the above-described intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion are implemented.

[0033] This invention provides an intelligent fault diagnosis method, system, electronic device, and storage medium for reciprocating compressor indicator diagrams based on multi-feature fusion. By deeply mining the rich physical information contained in the reciprocating compressor indicator diagram, a multi-dimensional feature space is constructed. An adaptive weight allocation strategy is used to achieve feature fusion, and a precise fault level determination mechanism is established based on fuzzy logic theory. This invention achieves high-precision, multi-fault, and refined intelligent diagnosis of reciprocating compressor indicator diagrams, improves the overall fault identification accuracy, effectively reduces false alarms and missed alarms, and provides a reliable basis for equipment maintenance. Attached Figure Description

[0034] Figure 1 A flowchart of an intelligent fault diagnosis method for a reciprocating compressor based on multi-feature fusion is provided by the present invention.

[0035] Figure 2 A schematic diagram of an LSTM-Attention hybrid network structure provided for one embodiment;

[0036] Figure 3 Figures (a) to (d) are comparative diagrams of the results of fault identification using four methods in a certain experimental verification scenario;

[0037] Figure 4 This is a comparison chart of the accuracy, recall, and F1 score of four methods for fault identification in a certain experimental verification scenario.

[0038] Figure 5 Figures (a) to (d) are comparative diagrams showing the results of identifying five types of fault data and health data using four methods in a certain experimental verification scenario.

[0039] Figure 6 This is a radar chart used in an experimental verification scenario to identify five types of fault data and health data using four methods.

[0040] Figure 7 A block diagram of a reciprocating compressor indicator diagram intelligent fault diagnosis system based on multi-feature fusion provided by the present invention;

[0041] Figure 8 A schematic diagram of a possible hardware structure of an electronic device provided by the present invention;

[0042] Figure 9 This is a schematic diagram of the hardware structure of a possible computer-readable storage medium provided by the present invention. Detailed Implementation

[0043] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0044] First, we will introduce the theoretical basis and technical principles upon which this invention is based.

[0045] The theoretical basis of this invention is built upon the following key theories:

[0046] 1. Thermodynamic Cycle Theory of Reciprocating Compressors

[0047] The working process of a reciprocating compressor follows the first and second laws of thermodynamics, and its ideal cycle can be described by polytropic process equations: = constant, where This is a variable index. In actual operation, due to factors such as friction loss, heat transfer loss, and leakage loss, the actual indicator diagram will deviate from the ideal cycle. Different types of faults will lead to specific deviation patterns, which provides a theoretical basis for fault identification.

[0048] 2. Signal Processing and Feature Extraction Theory

[0049] A dynamometer diagram is essentially a two-dimensional signal containing rich frequency and time domain information. By applying digital signal processing techniques, such as filtering, interpolation, and transform, fault characteristic signals hidden in noise can be effectively extracted. Furthermore, based on geometric shape analysis theory, fault information can be extracted from the geometric features of the dynamometer diagram, such as its area, perimeter, and key point locations.

[0050] 3. Pattern Recognition and Machine Learning Theory

[0051] Fault diagnosis is essentially a pattern recognition problem, namely, identifying different fault modes from a multi-dimensional feature space. This invention employs a combination of supervised and unsupervised learning methods, establishing a fault mode library through training on historical data, and using an adaptive algorithm to achieve online learning and model updates.

[0052] 4. Fuzzy Logic and Uncertainty Reasoning Theory

[0053] In engineering practice, fault states often exhibit fuzziness and uncertainty, making it difficult for traditional hard-decision methods to accurately describe such fuzzy boundaries. This invention introduces fuzzy logic theory to establish a fuzzy membership function, enabling soft-decision and confidence assessment.

[0054] Based on the above theoretical foundation and technical principles, such as Figure 1 As shown in the flowchart, this embodiment of the invention provides an intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion. The method specifically includes the following steps:

[0055] S1, Obtain the indicator diagram of the reciprocating compressor and perform preprocessing;

[0056] S2, extract multi-dimensional features from the indicator diagram of the reciprocating compressor;

[0057] S3, the multi-dimensional features are fused based on an adaptive weight allocation strategy and a comprehensive fault score is calculated;

[0058] S4, diagnose the fault type based on the feature combination and weight configuration in the fused features;

[0059] S5. Based on the comprehensive fault score, the fault level is determined by combining fuzzy boundary processing and dynamic threshold adaptive update mechanism.

[0060] Understandably, given the shortcomings in the background technology, this invention proposes an intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion. This method employs a layered and progressive technical architecture, mainly comprising five core layers: a data preprocessing layer, a feature extraction layer, a feature fusion layer, an intelligent diagnosis layer, and a fault classification result output layer. This invention deeply mines the rich physical information contained in the reciprocating compressor indicator diagram, constructing a multi-dimensional feature space including energy, dynamics, geometry, loss, and proportion. It employs an adaptive weight allocation strategy to achieve feature fusion and establishes a precise fault level determination mechanism based on fuzzy logic theory.

[0061] This invention achieves high-precision, multi-fault, and refined intelligent diagnosis of reciprocating compressor indicator diagrams. Experiments have verified that its comprehensive identification accuracy rate is over 95%, which is significantly better than traditional methods. It effectively reduces false alarms and missed alarms, and provides a reliable basis for equipment maintenance.

[0062] In one possible embodiment, step S1 involves obtaining the indicator diagram of the reciprocating compressor and performing preprocessing. The preprocessing process employs a multi-level preprocessing strategy, including sub-steps S101 to S105.

[0063] S101, Data Acquisition and Synchronization: Acquire dynamometer data during the operation of the reciprocating compressor, wherein the dynamometer data includes at least pressure data. and displacement data and the pressure data and displacement data Perform time alignment to ensure the time synchronization of the two signals, with a synchronization accuracy requirement of less than 1ms;

[0064] For example, real-time data acquisition is performed using a 1000Hz sampling frequency, ±0.1%FS sensor accuracy, and <1ms synchronization error.

[0065] S102, Adaptive Filtering: Considering the periodicity of the dynamometer card data, a circular smoothing filter algorithm is used to process the pressure data. and displacement data By processing the signal, this algorithm can effectively maintain the periodicity of the signal while eliminating high-frequency noise interference.

[0066] For example, the parameters of the circular smoothing filter algorithm are set as follows: window length 15-25, polynomial order 2-4, and periodic boundary expansion.

[0067] S103, Intelligent Outlier Detection: Based on the improved Z-score and interquartile range methods, it automatically identifies and corrects pressure data. and displacement data Outliers in the data should be identified to avoid the impact of sensor malfunctions or electromagnetic interference on diagnostic results.

[0068] Specifically, the improved Z-score method and interquartile range method include:

[0069] Use robust statistics instead of the mean and standard deviation, including replacing the mean with the median, replacing the standard deviation with the interquartile range (IQR), and setting the correction threshold to 3.5.

[0070] S104, High-precision interpolation compensation: for pressure data and displacement data For sampling points with missing or poor-quality data, cubic spline interpolation or shape-preserving piecewise cubic interpolation methods are used for compensation to ensure the integrity and continuity of the data.

[0071] S105, Standardized Index Establishment: Create a standardized displacement subscript system to map dynamometer diagram data under different working conditions and speeds to a unified coordinate system, laying the foundation for subsequent feature extraction and comparative analysis.

[0072] It is understandable that data preprocessing is the foundation of the entire diagnostic system, and its quality directly affects the accuracy of subsequent feature extraction and fault identification. This embodiment constructs a multi-level intelligent preprocessing system based on the above sub-steps, and also constructs a comprehensive data quality evaluation system to evaluate the signal-to-noise ratio (SNR), integrity, and consistency indicators.

[0073] The dynamometer diagram, as a graphical representation of the compressor's working cycle, shows that every subtle change corresponds to a specific physical process change within the equipment. Feature extraction in step S2 is the core of fault diagnosis. Based on thermodynamic theory and fluid mechanics principles, as well as the working principle and fault mechanism of reciprocating compressors, it extracts multi-dimensional core feature parameters with clear physical meaning from the dynamometer diagram. Each dimension has unique physical meaning and fault indication capabilities. These feature parameters not only reflect the current operating status of the equipment but, more importantly, reveal potential fault development trends.

[0074] For example, these core feature dimensions are:

[0075] 1. Energy characteristics dimension: The energy efficiency and working capacity of the compressor are reflected by the area of ​​the indicator diagram;

[0076] 2. Dynamic characteristics dimension: Pressure gradient analysis reflects valve action and fluid flow characteristics;

[0077] 3. Geometric characteristic dimension: The ideality of reflecting the thermodynamic process is evaluated through linear fitting;

[0078] 4. Loss Feature Dimension: Energy loss and efficiency reduction are quantified by calculating the loss area;

[0079] 5. Proportional characteristic dimension: The change of basic thermodynamic parameters is reflected through compression ratio analysis.

[0080] Based on the above-mentioned feature extraction theory, in one possible embodiment, step S2, extracting multi-dimensional features from the indicator diagram of the reciprocating compressor, includes sub-steps S201~S205:

[0081] S201, calculate the area of ​​the closed curve on the indicator diagram to obtain the energy characteristics.

[0082] The area of ​​the indicator diagram is an important indicator reflecting the energy consumption of the compressor's working cycle, directly related to the equipment's operating efficiency and fault status. In this embodiment, the area of ​​the closed curve of the indicator diagram can be calculated using the trapezoidal integral method:

[0083]

[0084] in: The area of ​​the indicator diagram is expressed in units of 1. This represents the net work done within one work cycle. Indicates the first The pressure values ​​at each sampling point are in Pa, and the range is typically 0.1-2.0 MPa. Indicates the first The displacement value corresponding to each sampling point, in meters, represents the relative position of the piston; This represents the total number of data points in a complete work cycle, typically between 1000 and 4000 points. This indicates the sampling point number, starting from 1.

[0085] This trapezoidal integral formula is based on the principle of numerical integration. It discretizes the continuous curve into several trapezoids and sums the areas of each trapezoid to obtain the total area. Under normal operating conditions, the area on the dynamometer card should be maintained within ±5% of the design value; exceeding this range indicates a fault.

[0086] S202, dynamic characteristics were obtained through pressure gradient analysis.

[0087] The pressure gradient reflects the drastic nature of pressure changes and is a key parameter for identifying valve malfunctions and sealing performance. In this embodiment, the central difference method is used to calculate the pressure change rate for each operating phase:

[0088]

[0089] in: This represents the pressure gradient, measured in Pa / s, and indicates the rate of change of pressure over time. This represents the pressure value at the next time step after the current moment, in Pa. This represents the pressure value at the previous time step, expressed in Pa. This represents a time interval, measured in seconds (s), and is typically 1-3 times the sampling period. This indicates the current time, expressed in seconds (s).

[0090] Different gradient thresholds are set for different operating phases:

[0091] Inspiratory phase: (Normal range)

[0092] Compressed phase: (Normal range)

[0093] Exhaust phase: (Normal range)

[0094] Expansion phase: (Normal range)

[0095] Abnormal pressure gradient changes typically indicate malfunctions such as valve sticking, seal leaks, or pipe blockages.

[0096] S203, geometric features were obtained through linear fitting evaluation.

[0097] Ideally, the compression and expansion processes should follow polytropic processes. Linear fitting analysis of deviations can effectively identify cylinder sealing performance and valve operating conditions. In this embodiment, the least squares method is used for linear fitting, and the goodness of fit is calculated:

[0098]

[0099] in: represents the goodness-of-fit coefficient, which is dimensionless and ranges from [0,1]. Indicates the first Each actual pressure measurement value is in Pa. Indicates the first The fitted pressure values ​​were calculated using a linear regression equation, and the unit is Pa. This represents the arithmetic mean of all actual pressure values, expressed in Pa.

[0100] The linear fitting equation is:

[0101]

[0102] in: This represents the slope of the fitted straight line, reflecting the trend of pressure change; This represents the intercept of the fitted line; Indicates the independent variable (displacement or time).

[0103] Goodness-of-fit evaluation criteria:

[0104] The linear relationship is good, and the equipment is working normally.

[0105] Slight deviation, requires attention.

[0106] Significant deviation, indicating a malfunction.

[0107] Significant deviation, equipment malfunction.

[0108] This method is particularly suitable for detecting faults that affect the compression and expansion process, such as abnormal cylinder clearance volume and piston ring wear.

[0109] S204, loss characteristics are obtained through loss area calculation.

[0110] Energy loss area quantifies the deviation between the actual working cycle and the ideal cycle, and is an important indicator for assessing the degree of equipment performance degradation. In this embodiment, energy loss is calculated by comparing the area difference between the actual indicator diagram and the ideal baseline:

[0111]

[0112] in: Indicates the area of ​​loss, in units of , representing the energy loss per unit cycle; The area of ​​the dynamometer card under ideal operating conditions is represented by units of 1. ; This represents the actual measured area of ​​the indicator diagram, in units of... .

[0113] Ideal area calculation is based on equipment design parameters:

[0114]

[0115] The phase loss area calculation includes the intake loss area calculation and the exhaust loss area calculation, specifically:

[0116] 1. Inhalation loss area:

[0117]

[0118] 2. Exhaust loss area:

[0119]

[0120] In the formula for calculating the area of ​​phase loss, This indicates crankshaft rotation angle, in degrees; This represents the ideal inspiratory pressure, which is usually equal to the inspiratory duct pressure. This represents the ideal exhaust pressure, which is usually equal to the exhaust pipe pressure.

[0121] Evaluation criteria for lost area:

[0122] <5%× : Normal operating conditions;

[0123] 5%× ≤ <15%× Minor abnormality;

[0124] 15%× ≤ <30%× Medium-level fault;

[0125] ≥30%× Serious malfunction.

[0126] S205, through compression ratio analysis, yielded proportional characteristics.

[0127] Compression ratio is a fundamental parameter reflecting the operating characteristics of a compressor, and its deviation directly indicates cylinder sealing performance and valve operating status. In this embodiment, equipment performance is evaluated by comparing the deviation between the actual compression ratio and the theoretical design value.

[0128]

[0129] in: It represents the relative deviation of the compression ratio, is dimensionless, and indicates the degree of deviation; This represents the actual compression ratio and is dimensionless. This represents the theoretical compression ratio, is dimensionless, and is determined by the equipment design parameters.

[0130] Actual compression ratio calculation:

[0131]

[0132] in: This indicates the maximum pressure value within one work cycle, expressed in Pa. This indicates the minimum pressure value within one work cycle, expressed in Pa.

[0133] Theoretical compression ratio calculation:

[0134]

[0135] in: This indicates the piston scavenging volume, in cubic meters (m³). 3 ; This indicates the cylinder clearance volume, in meters (m). 3 .

[0136] Corrected compression ratio considering polytropic processes:

[0137]

[0138] in: This indicates the exhaust pressure, expressed in Pa. This represents the inhalation pressure, expressed in Pa. This represents the variability index, typically ranging from 1.2 to 1.4.

[0139] Compression ratio deviation evaluation criteria:

[0140] <0.05: Compression ratio is normal;

[0141] 0.05≤ <0.15: Slight deviation, requires monitoring;

[0142] 0.15≤ <0.30: Significant deviation, indicating a malfunction;

[0143] ≥0.30: Severe deviation, equipment malfunction.

[0144] An abnormal compression ratio usually indicates problems such as excessive cylinder clearance, worn piston rings, or valve leakage.

[0145] In one possible embodiment, step S3 involves fusing the multidimensional features based on an adaptive weight allocation strategy and calculating a comprehensive fault score, specifically including sub-steps S301 to S303.

[0146] S301, based on historical prior data, configures basic weights, sensitivity coefficients and fault indicator logic for features in each dimension according to different fault types.

[0147] It is understandable that the basic weight allocation strategy involves configuring corresponding basic weights for different fault types, for example:

[0148] a) For faults involving excessive cylinder clearance volume, the basic weighting configuration is as follows:

[0149] (Area of ​​the indicator diagram) = 0.30; (Pressure gradient) = 0.20; (Linear fit) = 0.25; (Loss area) = 0.15; (Compression ratio) = 0.10.

[0150] b) For valve failures, the basic weight configuration is as follows:

[0151] (Area of ​​the indicator diagram) = 0.20; (Pressure gradient) = 0.35; (Linear fit) = 0.15; (Loss area) = 0.20; (Compression ratio) = 0.10.

[0152] The formula for determining the sensitivity coefficient is:

[0153]

[0154] The sensitivity coefficient is determined by the following formula: This represents the feature sensitivity benchmark coefficient, with a value ranging from 0.2 to 0.4. Indicates the first Historical standard deviation of each feature; Indicates the first The historical mean of each feature.

[0155] The logical expression for the fault indicator is:

[0156]

[0157] In the fault indicator logic expression, Indicates the first The baseline values ​​for each feature under normal operating conditions Indicates the degree of deviation of features. For the first The anomaly detection threshold for each feature. The fault indicator is triggered when the deviation of a feature from its corresponding anomaly detection threshold exceeds the threshold value; in this case, the feature triggers the fault warning mode.

[0158] S302, based on the basic weights, sensitivity coefficients and fault indicator logic, an adaptive weight allocation algorithm is used to calculate the adaptive weight coefficients of the corresponding features in each dimension of the multi-dimensional features, and the weights of each dimension in a single fault type are normalized.

[0159] The adaptive weight calculation formula is:

[0160]

[0161] in: Indicates the first The basic weight coefficients of each feature are determined based on the fault type and feature importance; Indicates the first The sensitivity adjustment coefficient for each feature, with a value range of [0.1, 0.5]; Indicates the first The baseline value of each feature under normal operating conditions; It indicates the degree of deviation of the feature and reflects the severity of the anomaly.

[0162] Weight normalization formula:

[0163]

[0164] In the weight normalization formula, The weights are the normalized weights for the i-th feature. Let be the original weight of the i-th feature before normalization, given by an adaptive weight allocation algorithm or expert experience / basic weight configuration, and j be the summation variable. It is the sum of the weights of m features.

[0165] Weight normalization ensures that the sum of all weight coefficients equals 1, guaranteeing the consistency and comparability of the scoring results.

[0166] S303, based on the adaptive weight coefficients of the corresponding features of each dimension, fuses multi-dimensional features and calculates the comprehensive fault score.

[0167] In this embodiment, by integrating multi-dimensional feature information and combining it with an adaptive weight allocation algorithm to calculate a comprehensive fault score, quantitative fault assessment can be achieved. The formula for calculating the comprehensive fault score is:

[0168]

[0169] In the formula for calculating the comprehensive fault score, This represents the overall fault score, i.e., the total fault score. It is dimensionless and ranges from [0, 100]. Indicates the number of digits after normalization. The adaptive weight coefficients of each feature are dimensionless and satisfy the following conditions: ; Indicates the first The standardized eigenvalues ​​of each feature, dimensionless, with a value range of [0,1]; Indicates the first A fault indicator with a characteristic, taking a value of 0 or 1; This represents the total number of features involved in the scoring. In this embodiment, there are 5 core features, so m=5. This indicates the feature number, starting from 1.

[0170] In one possible embodiment, step S4, configuring the fault type based on the feature combination and weights in the fused features, includes:

[0171] 1. Establish and continuously update the fault mode library based on historical data of indicator diagrams under multiple operating conditions and speeds.

[0172] Understandably, the fault mode library establishes specialized fault diagnosis models for different fault types, for example:

[0173] 1) Fault diagnosis model for excessive cylinder clearance volume;

[0174] 2) Intake valve fault diagnosis model;

[0175] 3) Fault diagnosis model for exhaust valve seizure;

[0176] 4) Fault diagnosis model for non-leaky intake and exhaust valves;

[0177] 5) Piston ring leakage fault diagnosis model;

[0178] 6) Fault diagnosis model for exhaust valve plate tripping;

[0179] 7) Exhaust pipe blockage fault diagnosis model.

[0180] Each fault type employs a specific combination of features and weight configuration to facilitate accurate fault type identification during the detection process.

[0181] 2. If the standardized feature value of any feature in the current fusion features is... Trigger fault indicator After traversing all features, count all fault features that trigger the fault indicator in the current fused features, and obtain the adaptive weight coefficients corresponding to each fault feature.

[0182] 3. Match all fault features and their corresponding adaptive weight coefficients with the fault mode library, and find the fault type with the highest similarity based on the preset fault diagnosis models.

[0183] After determining the fault type in step S4, the severity of the fault can be further classified in step S5. In one possible embodiment of step S5, the fault level is determined based on the comprehensive fault score, combined with fuzzy boundary processing and a dynamic threshold adaptive update mechanism, including sub-steps S501 to S504.

[0184] S501, Set multi-level fault judgment thresholds to form multi-level fault intervals, and determine the fault level according to the fault interval into which the comprehensive fault score falls.

[0185] The precise logical expression for classifying fault severity is as follows:

[0186]

[0187] in, The fault level is an integer value, divided into 4 levels. A value of 0 indicates a normal state, meaning the equipment is operating well. A score of 1 indicates a minor fault that requires attention but can continue to operate. A fault rating of 2 indicates a moderate fault, requiring a maintenance plan to be arranged. A value of 3 indicates a serious fault, requiring immediate shutdown and repair. This is the overall fault score, with a value range of [0, 100]. The threshold for minor faults is set to 25 based on experience. The threshold for determining a medium-level fault is set to 55 based on experience. The initial value for the severe fault determination threshold is set to 80 based on experience.

[0188] S502, for the comprehensive fault score located near the fault determination threshold, a membership function is introduced to perform boundary fuzzy judgment, and the fault level determination result is corrected based on the membership degree of the comprehensive fault score belonging to each level of fault determination interval.

[0189] To address the fuzzy boundary handling, a membership function is introduced to handle the fuzziness of the threshold boundary:

[0190]

[0191] In the membership function, Overall Fault Score Belongs to the The membership degree of the level fault, with a value of [0,1]; The width of the fuzzy boundary is empirically set to 15% of the difference between adjacent thresholds.

[0192] S503, taking into account the data distribution characteristics, assesses the confidence level of the corrected fault level determination result.

[0193] The confidence level assessment formula is as follows:

[0194]

[0195] in: The confidence level of the fault level diagnosis result, with a value range of [0,1]. The characteristic coefficient of variation, ; The standard deviation of all eigenvalues; This is the mean of all eigenvalues.

[0196] S504 pre-sets a dynamic threshold adaptive update function and a threshold boundary protection function. Based on the current comprehensive fault score, historical data, and confidence level, it uses deep learning methods to learn the trend of fault judgment threshold changes in order to dynamically adjust the fault judgment threshold.

[0197] The dynamic threshold adaptive update function is expressed as:

[0198]

[0199] in: For the first Updated after the first diagnosis Level threshold; The first diagnosis Level threshold; The learning rate parameter, with a value range of [0.01, 0.05], controls the speed of threshold adjustment; The change in score .

[0200] The threshold boundary protection function is:

[0201]

[0202] in and The first The minimum and maximum values ​​of the threshold can be set, and a threshold boundary protection mechanism can be set to prevent misjudgment caused by excessive threshold adjustment.

[0203] A deep learning model is constructed by combining fuzzy boundary processing and dynamic threshold adaptive update mechanism to achieve self-learning of the fault judgment threshold change trend using deep learning methods, and then dynamically adjust the fault judgment threshold.

[0204] like Figure 2 As shown, a deep learning model is illustrated using an LSTM-Attention hybrid network. This LSTM-Attention hybrid network includes an input layer, an LSTM layer, an Attention layer, a fully connected layer, and an output layer. The specific parameter settings are as follows:

[0205] Input layer: Input 3D vector [ , , ],in This is the overall fault score obtained from the previous diagnosis. This represents the difference between the overall fault score of this diagnosis and the previous diagnosis. The confidence level of the previous diagnosis;

[0206] LSTM layer: 64 units The LSTM layer, acting as a self-memory temporal feature encoder, provides physically meaningful temporal semantics for the Attention layer.

[0207] Attention layer: Acts as an automatic weight assigner, calculating the weights of the historical sequence using the following formula:

[0208] ,

[0209] in, Let be the attention weight vector for the hidden states at each time step t in the historical sequence, representing the importance allocation of the current time step to the information of each past time step; This is the output weight matrix in the attention mechanism, used to map the hidden states to the attention score space, with dimensions of [missing information]. ,in For the hidden layer dimension of the LSTM, For attention hidden layer dimensions; This is the hidden state transformation weight matrix in the attention mechanism, used to perform linear transformations on historical hidden states to enhance semantic expressiveness. Its dimension is... ; This is the hidden state output of the LSTM at time step t, containing the context information and long-term dependent memory at that moment, with a dimension of [missing information]. ; This refers to the bias term in the hidden state transition of the attention mechanism, with dimension . This is used to adjust the input offset of the activation function.

[0210] Fully connected layer: Serves as the decision layer, 2 layers (32→16 neurons), using the ReLU activation function to map the context vector extracted from the Attention layer into 3 physical quantities that can be directly used to update the threshold. ;

[0211] Output layer: Outputs a 3D vector , That is, the updated fault determination threshold.

[0212] Finally, the output format of the level determination by the method of this invention is as follows:

[0213]

[0214] To verify the method of the present invention, an example of an experimental analysis process is provided below.

[0215] The indicator diagram data under both normal and fault conditions were selected for compressor fault identification. First, the compressor indicator diagram signals under the same speed and load conditions were extracted at equal time intervals. The signals for each state were divided into 200 groups, that is, each state contained 200 samples, as shown in Table 1.

[0216] Table 1 Experimental Dataset

[0217]

[0218] To verify the effectiveness of the proposed method, it was compared with single-feature methods, fixed-weight methods, and binary decision methods. The dataset was divided into an 80% training set and a 20% test set. The fault identification results of each method are presented using a line mixing matrix, as shown below. Figure 3 As shown in (a) to (d). Compare. Figure 3 As shown in the results (a) to (d), the accuracy of the method proposed in this invention is as high as 98.75%, which is 12.5%, 16.25%, and 21.25% higher than the other three methods, respectively. This verifies the effectiveness of the method and proves its superiority in compressor fault diagnosis. Figure 4 To compare the precision, recall, and F1 score (the harmonic mean of precision and recall) of the four methods, by... Figure 4 As can be seen, the accuracy, recall, and F1 score of the method proposed in this invention are significantly higher than those of the other three methods, which verifies the correctness and effectiveness of the intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-dimensional feature fusion in compressor fault diagnosis, and is of great significance to the normal operation of the compressor.

[0219] To verify the universality of the proposed method, six compressor indicator diagram data sets were used, including five fault types (rod breakage, fixed valve leakage, upper piston leakage, insufficient liquid supply, and lower piston leakage) plus health data, to validate the method. Each compressor indicator diagram data set contained 200 data points, divided into training and test sets in a 7:3 ratio, as shown in Table 2. The proposed method was compared with the single-feature method, fixed-weight method, and binary decision method, and the results are as follows: Figure 5 As shown in (a) to (d).

[0220] Table 2 Experimental Dataset Division

[0221]

[0222] like Figure 5As shown in (a) to (d), the accuracy of the method proposed in this invention reaches 95.28%, with only 8 false predictions in the identification of category 4 faults (i.e., lower piston leakage faults) in Table 2. However, this is still superior to the other three methods. The single-feature method had 3 false predictions in category 4, the fixed-weight method had 15 false predictions, and the binary decision method had 20 false predictions. Furthermore, the diagnostic accuracy of the method proposed in this invention is also superior to the other three methods in other categories. The overall accuracy rates of the other three methods are 83.33%, 79.17%, and 71.94%, respectively, all lower than the accuracy rate of the method proposed in this invention, thus further demonstrating the effectiveness of the proposed method in compressor fault diagnosis. Figure 6 As can be seen from the radar chart, the method proposed in this invention has a higher diagnostic accuracy than the other three methods in all five types of fault and health data shown in Table 2. These verifications demonstrate the effectiveness of the method proposed in this invention in diagnosing compressor indicator diagrams.

[0223] Figure 7 This invention provides a structural diagram of an intelligent fault diagnosis system for reciprocating compressor indicator diagrams based on multi-feature fusion, as shown in the embodiment of the invention. Figure 7 As shown, an intelligent fault diagnosis system for reciprocating compressors based on multi-feature fusion indicator diagrams includes an acquisition module, a weight allocation and scoring module, a fault type diagnosis module, and a fault classification module, wherein:

[0224] The acquisition module is used to extract multi-dimensional features from the indicator diagram of the reciprocating compressor;

[0225] The weight allocation and scoring module is used to fuse the multi-dimensional features based on an adaptive weight allocation strategy and calculate a comprehensive fault score.

[0226] The fault type diagnosis module is used to diagnose fault types based on the feature combinations and weights in the fused features.

[0227] The fault classification module is used to determine the fault level based on the comprehensive fault score, combined with fuzzy boundary processing and dynamic threshold adaptive update mechanism.

[0228] It is understood that the intelligent fault diagnosis system for reciprocating compressor indicator diagrams based on multi-feature fusion provided by this invention corresponds to the intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion provided in the foregoing embodiments. The relevant technical features of the intelligent fault diagnosis system for reciprocating compressor indicator diagrams based on multi-feature fusion can be referred to the relevant technical features of the intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion, and will not be repeated here.

[0229] Please see Figure 8 , Figure 8 This is a schematic diagram illustrating an embodiment of the electronic device provided in this invention. For example... Figure 8 As shown, this embodiment of the invention provides an electronic device 800, including a memory 810, a processor 820, and a computer program 811 stored in the memory 810 and executable on the processor 820. When the processor 820 executes the computer program 811, it performs the following steps:

[0230] Extracting multidimensional features from the indicator diagram of a reciprocating compressor;

[0231] The multidimensional features are fused based on an adaptive weight allocation strategy, and a comprehensive fault score is calculated.

[0232] The fault type is diagnosed based on the feature combination and weight configuration in the fused features;

[0233] Based on the comprehensive fault score, the fault level is determined by combining fuzzy boundary processing and dynamic threshold adaptive update mechanism.

[0234] Please see Figure 9 , Figure 9 This is a schematic diagram illustrating an embodiment of a computer-readable storage medium provided by the present invention. (See diagram below.) Figure 9 As shown, this embodiment provides a computer-readable storage medium 900, on which a computer program 911 is stored. When the computer program 911 is executed by a processor, it performs the following steps:

[0235] Extracting multidimensional features from the indicator diagram of a reciprocating compressor;

[0236] The multidimensional features are fused based on an adaptive weight allocation strategy, and a comprehensive fault score is calculated.

[0237] The fault type is diagnosed based on the feature combination and weight configuration in the fused features;

[0238] Based on the comprehensive fault score, the fault level is determined by combining fuzzy boundary processing and dynamic threshold adaptive update mechanism.

[0239] The present invention provides an intelligent fault diagnosis method, system, and storage medium for reciprocating compressor indicator diagrams based on multi-feature fusion, which has the following advantages compared to the prior art:

[0240] 1. Systemic advantages of multidimensional feature fusion

[0241] This invention overcomes the limitations of traditional single-feature diagnosis and establishes a comprehensive feature system encompassing five dimensions: energy, dynamics, geometry, loss, and scale. This multi-dimensional feature fusion method has the following advantages:

[0242] a) Information integrity: Fully capture the physical information in the indicator diagram and avoid the one-sidedness of a single feature;

[0243] b) Enhanced complementarity: Different feature dimensions complement each other, improving the robustness of fault identification;

[0244] c) Redundancy guarantee: Parallel analysis of multiple features ensures that even if one feature is disturbed, other features can still guarantee the reliability of the diagnosis.

[0245] d) Clear physical meaning: Each feature corresponds to a clear physical process, which is convenient for engineers to understand and verify.

[0246] Extensive engineering practice has verified that, compared with single-feature methods, the multi-dimensional feature fusion method improves the fault identification accuracy by 12.5% ​​and significantly reduces the false alarm rate.

[0247] 2. The innovation of adaptive intelligent weight allocation

[0248] The adaptive weight allocation algorithm proposed in this invention is a significant improvement over the traditional fixed weight method:

[0249] a) Dynamic adaptability: The weighting coefficients can be automatically adjusted according to the intensity of the actual fault characteristics, avoiding the subjectivity of manually setting the weights;

[0250] b) Fault sensitivity: For different fault types, the system can automatically highlight the most sensitive characteristic parameters, improving the pertinence of diagnosis;

[0251] c) Learning ability: Through the accumulation of historical data, the weight allocation strategy is continuously optimized to achieve continuous improvement in system performance;

[0252] d) Adaptability to operating conditions: It can adapt to changes in the importance of features under different operating conditions, ensuring the consistency of diagnostic results.

[0253] Experimental results show that, compared with the fixed weight method, the adaptive weight allocation improves the diagnostic accuracy by 16.25% and significantly enhances its adaptability to complex working conditions.

[0254] 3. Accuracy of fuzzy logic level determination

[0255] Traditional hard-decision methods are prone to misjudgment at fault boundaries. This invention solves this problem by using fuzzy logic theory:

[0256] a) Soft decision mechanism: Soft decision is achieved through membership function, which avoids the sudden change problem caused by hard threshold;

[0257] b) Confidence assessment: Provide a confidence index for each diagnostic result to facilitate engineers in assessing the reliability of the diagnosis;

[0258] c) Dynamic threshold adjustment: The threshold can be automatically adjusted according to the equipment's operating history to adapt to equipment aging and changes in operating conditions;

[0259] d) Refined classification: Achieve a refined fault level classification of 0-3, providing a more accurate basis for maintenance decisions.

[0260] Compared to traditional binary decision-making methods, the accuracy of fuzzy logic level determination is improved by more than 20%, especially in the identification ability of the initial stage of fault and boundary state.

[0261] 4. Comprehensiveness of multi-fault type identification

[0262] This invention establishes a unified diagnostic framework covering a variety of common fault types:

[0263] a) Comprehensive fault coverage: It covers major fault types such as rod breakage, valve failure, piston ring wear, and insufficient fluid supply;

[0264] b) Specialized diagnostic models: Each fault type has a specially optimized diagnostic model to ensure accurate identification;

[0265] c) Strong fault differentiation capability: It can effectively distinguish similar fault types and avoid misdiagnosis and missed diagnosis;

[0266] d) High recognition accuracy: The overall recognition accuracy reaches over 95%, and the recognition accuracy for a single fault type can reach over 98%.

[0267] Through application verification in multiple industrial sites, this invention has significantly improved the accuracy and efficiency of equipment fault diagnosis, saving enterprises a large amount of maintenance costs and downtime losses, and has significant economic value and social benefits.

[0268] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0269] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0270] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0271] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0272] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0273] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0274] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent fault diagnosis of reciprocating compressors based on multi-feature fusion indicator diagrams, characterized in that, include: Extracting multidimensional features from the indicator diagram of a reciprocating compressor; The multidimensional features are fused based on an adaptive weight allocation strategy to calculate a comprehensive fault score, including: Based on historical prior data, configure basic weights, sensitivity coefficients, and fault indicator logic for features in each dimension according to different fault types; Based on the aforementioned basic weights, sensitivity coefficients, and fault indicator logic, an adaptive weight allocation algorithm is used to calculate the adaptive weight coefficients of each dimension of the multidimensional features, and the weights of each dimension in a single fault type are normalized. The multi-dimensional features are fused based on the adaptive weight coefficients of the corresponding features in each dimension, and the comprehensive fault score is calculated. The calculation formula is as follows: in, For the comprehensive fault score, For the normalized version The adaptive weight coefficients of each feature satisfy... =1, Indicates the first Standardized feature values ​​of each feature Indicates the first A fault indicator with specific characteristics. Take 0 or 1, The feature index is represented as i∈[1,m]. This represents the total number of features involved in the scoring; Fault indicator The logical representation is as follows: in, Indicates the first The baseline values ​​for each feature under normal operating conditions Indicates the degree of deviation of the feature. For the first The anomaly detection threshold for each feature; Based on the feature combinations and weight configurations in the fused features, fault types are diagnosed, including: A fault mode library is established and continuously updated based on historical data of indicator diagrams under multiple operating conditions and speeds. If the standardized feature value of any feature in the current fusion features Trigger fault indicator Count all fault features that trigger fault indicators in the current fused features, and obtain the adaptive weight coefficients corresponding to each fault feature; All fault features and their corresponding adaptive weight coefficients are matched with the fault pattern library to find the fault type with the highest similarity. Based on the comprehensive fault score, the fault level is determined by combining fuzzy boundary processing and dynamic threshold adaptive update mechanism.

2. The intelligent fault diagnosis method for reciprocating compressors based on multi-feature fusion according to claim 1, characterized in that, The process also includes preprocessing the indicator diagram, the preprocessing comprising: Acquire dynamometer data during the operation of a reciprocating compressor, the dynamometer data including pressure data. and displacement data and the pressure data and displacement data Perform time alignment; A circular smoothing filter algorithm is used to eliminate pressure data. and displacement data High-frequency noise in; Based on the improved Z-score and interquartile range methods, pressure data is identified and corrected. and displacement data Outliers in; Cubic spline interpolation or conformal piecewise cubic interpolation methods are used to analyze the pressure data. and displacement data Compensation is provided for sampling points with missing or poor-quality data. The dynamometer diagram data under different operating conditions and speeds are mapped to a unified displacement subscript coordinate system.

3. The intelligent fault diagnosis method for reciprocating compressors based on multi-feature fusion indicator diagrams according to claim 2, characterized in that, The improved Z-score method and interquartile range method include: Use robust statistics instead of the mean and standard deviation, including replacing the mean with the median and the standard deviation with the interquartile range (IQR).

4. The intelligent fault diagnosis method for reciprocating compressors based on multi-feature fusion indicator diagrams according to claim 1, characterized in that, The extraction of multi-dimensional features from the indicator diagram of the reciprocating compressor includes: Calculate the area of ​​the closed curve on the indicator diagram to obtain the energy characteristics; Dynamic characteristics are obtained through pressure gradient analysis; Geometric features are obtained through linear fitting evaluation; The loss characteristics are obtained by calculating the loss area; The proportional characteristics were obtained through compression ratio analysis.

5. The intelligent fault diagnosis method for reciprocating compressors based on multi-feature fusion according to claim 1, characterized in that, The formula for calculating the adaptive weighting coefficient is: in: Indicates the first The basic weight coefficients of each feature are determined based on the fault type and feature importance; Indicates the first Sensitivity adjustment coefficient for each feature; Indicates the first The baseline value of each feature under normal operating conditions; It indicates the degree of deviation of the feature and reflects the severity of the anomaly; Sensitivity adjustment coefficient Represented as: in: This represents the baseline coefficient for feature sensitivity, which is determined empirically. Indicates the first Historical standard deviation of each feature; Indicates the first The historical mean of each feature.

6. The intelligent fault diagnosis method for reciprocating compressors based on multi-feature fusion indicator diagrams according to claim 1, characterized in that, The determination of the fault level based on the comprehensive fault score, combined with fuzzy boundary processing and a dynamic threshold adaptive update mechanism, includes: Set multi-level fault judgment thresholds to form multi-level fault intervals. Determine the fault level based on the fault interval into which the comprehensive fault score falls. For the comprehensive fault score located near the fault determination threshold, a membership function is introduced to perform boundary fuzzy judgment. Based on the membership degree of the comprehensive fault score belonging to each level of fault determination interval, the fault level determination result is corrected. Based on the data distribution characteristics, assess the confidence level of the corrected fault level determination results; A preset dynamic threshold adaptive update function and threshold boundary protection function are used. Based on the current comprehensive fault score, historical data and confidence level, a deep learning method is used to learn the trend of fault judgment threshold changes in order to dynamically adjust the fault judgment threshold.

7. A reciprocating compressor indicator diagram intelligent fault diagnosis system based on multi-feature fusion, characterized in that, include: The acquisition module is used to extract multi-dimensional features from the indicator diagram of the reciprocating compressor; The weight allocation and scoring module is used to fuse the multi-dimensional features based on an adaptive weight allocation strategy and calculate a comprehensive fault score, including: Based on historical prior data, configure basic weights, sensitivity coefficients, and fault indicator logic for features in each dimension according to different fault types; Based on the aforementioned basic weights, sensitivity coefficients, and fault indicator logic, an adaptive weight allocation algorithm is used to calculate the adaptive weight coefficients of each dimension of the multidimensional features, and the weights of each dimension in a single fault type are normalized. The multi-dimensional features are fused based on the adaptive weight coefficients of the corresponding features in each dimension, and the comprehensive fault score is calculated. The calculation formula is as follows: in, For the comprehensive fault score, For the normalized version The adaptive weight coefficients of each feature satisfy... =1, Indicates the first Standardized feature values ​​of each feature Indicates the first A fault indicator with specific characteristics. Take 0 or 1, The feature index is represented as i∈[1,m]. This represents the total number of features involved in the scoring; Fault indicator The logical representation is as follows: in, Indicates the first The baseline values ​​for each feature under normal operating conditions Indicates the degree of deviation of the feature. For the first The anomaly detection threshold for each feature; The fault type diagnosis module is used to configure fault type diagnosis based on feature combinations and weights in the fused features, including: A fault mode library is established and continuously updated based on historical data of indicator diagrams under multiple operating conditions and speeds. If the standardized feature value of any feature in the current fusion features Trigger fault indicator Count all fault features that trigger fault indicators in the current fused features, and obtain the adaptive weight coefficients corresponding to each fault feature; All fault features and their corresponding adaptive weight coefficients are matched with the fault pattern library to find the fault type with the highest similarity. The fault classification module is used to determine the fault level based on the comprehensive fault score, combined with fuzzy boundary processing and dynamic threshold adaptive update mechanism.

8. An electronic device, characterized in that, The method includes a memory and a processor, wherein the processor is used to execute computer management programs stored in the memory to implement the steps of the intelligent fault diagnosis method for reciprocating compressor indicator diagrams based on multi-feature fusion as described in any one of claims 1-6.

Images