Air compressor abnormality prediction method and system

Abstract

The present application relates to a kind of air compressor exception prediction method, system, comprising the following steps: S1, the multi-view feature vector of air compressor vibration is constructed;S2, causal relationship model is established, distinguish vibration characteristics driven by working condition variable from vibration characteristics driven by vibration underlying cause, and generate vibration mechanism prior image based on the causal relationship model;S3, the current vibration data of the air compressor is collected, the current vibration data is decomposed into the vibration estimation component corresponding to multiple vibration underlying causes by signal decomposition method, the vibration estimation component corresponding is constrained by the vibration mechanism prior image, and the vibration estimation component after optimization is obtained;S4, the performance of each vibration estimation component after optimization in the multi-view feature vector is evaluated, vibration cause contribution degree is assigned to each vibration estimation component after optimization, is sorted to form fault possibility judgment.

Description

Technical Field

[0001] This invention relates to the field of abnormal vibration monitoring of air compressors, specifically to a method and system for predicting abnormalities in air compressors. Background Technology

[0002] Air compressors are one of the main mechanical power equipment for many enterprises. With the continuous development of industrial level, industrial users have an increasingly strong demand for air compressors. The safe and stable operation of air compressors has also become an important issue for users. Among them, the vibration intensity of the main unit is an important indicator for assessing the health of the equipment and is used to analyze and evaluate the stability and reliability of the mechanical components of the equipment.

[0003] Air compressor vibration intensity is often used as a standard to evaluate the health of the air compressor main unit and drive motor. Furthermore, the real-time magnitude assessment of vibration intensity and the analysis of its trend changes are equally important in equipment health management. Current technology only determines whether the equipment's vibration intensity exceeds a threshold to decide whether maintenance is needed, but it has not yet achieved the accumulation of operating data and health trend prediction for air compressors based on their vibration intensity.

[0004] The purpose of this invention is to design an air compressor anomaly prediction method and system to address the problems existing in the prior art. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides an air compressor anomaly prediction method and system, which can effectively solve at least one of the problems existing in the prior art.

[0006] The technical solution of this invention is:

[0007] A method for predicting air compressor anomalies includes the following steps:

[0008] S1. Collect historical vibration data of the air compressor under different operating conditions, extract the acceleration perspective, velocity perspective, and displacement perspective of the air compressor from the historical vibration data, calculate the statistical characteristics of the acceleration perspective, velocity perspective, and displacement perspective respectively, and construct a multi-view feature vector of the air compressor vibration.

[0009] S2, collect the historical underlying causes of vibration of the air compressor and corresponding operating condition data, and construct a causal relationship model between operating condition variables, vibration feature variables and vibration underlying cause variables based on the multi-view feature vector, the operating condition data and the underlying causes of vibration, distinguish the vibration features driven by the operating condition variables and the vibration features driven by the underlying causes of vibration, and generate a priori profile of the vibration mechanism based on the causal relationship model.

[0010] S3, collect the current vibration data of the air compressor, decompose the current vibration data into multiple vibration prediction components corresponding to the underlying causes of vibration through the signal decomposition method, constrain the corresponding vibration prediction components through the vibration mechanism prior profile, and obtain the optimized vibration prediction components.

[0011] S4. Evaluate the performance of each optimized vibration prediction component in the multi-view feature vector, assign a vibration cause contribution to each optimized vibration prediction component, rank the vibration causes based on the vibration cause contribution, and form a fault probability judgment.

[0012] Further, in S1, extracting the acceleration perspective, velocity perspective, and displacement perspective of the air compressor from the historical vibration data includes: the historical vibration data includes an acceleration signal; the acceleration signal is high-pass filtered to obtain the acceleration perspective; the acceleration signal is integrated once and then band-pass filtered at a mid-frequency to obtain the velocity perspective; the acceleration signal is integrated twice and then low-pass filtered to obtain the displacement perspective.

[0013] The statistical characteristics include one or more of the following in terms of time and frequency: RMS, peak value, peak-to-peak value, variance, skewness, kurtosis, peak factor, impulse factor, and margin factor.

[0014] Furthermore, in step S2, by collecting historical maintenance work orders, maintenance behaviors are extracted from the historical maintenance work orders, and the underlying causes of the historical vibrations are inferred based on the maintenance behaviors;

[0015] The operating data includes one or more of the following: air compressor speed, load rate, exhaust pressure, temperature, and current.

[0016] Further, step S2 includes:

[0017] S2.1, construct a benchmark prediction model containing its own hysteresis term for each of the multi-view feature vectors based on the vibration feature variables, and construct an extended prediction model by adding the hysteresis term of the corresponding working condition variable to the benchmark prediction model.

[0018] S2.2, compare the prediction error difference between the benchmark prediction model and the extended prediction model. When the prediction error difference is greater than a preset threshold, establish the causal relationship edge between the corresponding working condition variable and the vibration characteristic variable and its causal strength value.

[0019] S2.3, Perform statistical differentiation analysis between vibration underlying cause variables and vibration characteristic variables under the same working conditions. When the difference in vibration characteristic distribution corresponding to different vibration underlying causes is greater than the preset differentiation threshold, establish the causal relationship edge from the vibration underlying cause variable to the vibration characteristic variable and its causal strength value.

[0020] S2.4 Based on the causal relationship edges and their causal strength values ​​established in steps S2.2 and S2.3, construct a causal relationship model that includes working condition variable nodes, vibration characteristic nodes, and vibration underlying cause nodes;

[0021] S2.5, Read the set of parent nodes and their causal intensity values ​​corresponding to each vibration feature node in the causal relationship model, calculate the sum of causal intensity from the working condition variable node for each vibration feature node to obtain the working condition driving intensity, calculate the sum of causal intensity from the vibration underlying cause node for each vibration feature node to obtain the mechanism driving intensity, when the mechanism driving intensity is greater than the working condition driving intensity, determine the vibration feature as a vibration feature driven by the vibration underlying cause, and generate a vibration mechanism prior profile.

[0022] Further, in step S3, the current vibration data is decomposed into multiple vibration prediction components corresponding to underlying causes of vibration using a signal decomposition method, including:

[0023] The current vibration data is decomposed into multiple intrinsic mode functions and a residual term using an integrated empirical mode algorithm.

[0024] Calculate the statistical and frequency domain characteristics of each intrinsic mode function, and determine the contribution of each intrinsic mode function to each vibration underlying cause based on the matching relationship between the statistical and frequency domain characteristics and the prior profile of the vibration underlying cause. Determine the vibration underlying cause corresponding to each intrinsic mode function based on the maximum contribution value.

[0025] By combining and reconstructing the intrinsic mode functions of the same underlying cause of vibration, the vibration prediction component of the underlying cause of vibration is obtained.

[0026] Furthermore, by combining and reconstructing the intrinsic mode functions of the same underlying cause of vibration, the vibration prediction components of that underlying cause are obtained, including:

[0027] The maximum contribution of each intrinsic mode function is normalized and used as the weight. The intrinsic mode functions of the same underlying cause of vibration are weighted and summed to obtain the vibration prediction component of the underlying cause of vibration.

[0028] Further, in step S3, the corresponding vibration prediction components are constrained by the prior profile of the vibration mechanism to obtain optimized vibration prediction components, including:

[0029] The underlying cause of vibration corresponding to the vibration prediction component is obtained, as well as the prior profile of the vibration mechanism corresponding to the underlying cause of vibration. Based on the prior profile of the vibration mechanism, the vibration prediction component is constrained in terms of frequency band, morphology and statistical characteristics. Components that do not conform to the mechanism are suppressed and components that conform to the mechanism are enhanced. The combined components are the current vibration data, thereby obtaining the optimized vibration prediction component.

[0030] Further, in step S4, the performance of each optimized vibration prediction component in the multi-view feature vector is evaluated, and the vibration cause contribution is assigned to each optimized vibration prediction component, including:

[0031] Calculate the energy percentage of the optimized vibration prediction component in the corresponding frequency band of the current vibration data, and use the energy percentage as the contribution of the vibration cause.

[0032] Furthermore, an air compressor anomaly prediction system is provided, comprising the following modules:

[0033] The multi-view feature construction module is used to collect historical vibration data of the air compressor under different operating conditions, extract the acceleration view, velocity view, and displacement view of the air compressor from the historical vibration data, calculate the statistical features of the acceleration view, velocity view, and displacement view respectively, and construct the multi-view feature vector of the air compressor vibration.

[0034] The vibration mechanism prior profile generation module is used to collect the historical underlying causes of vibration of the air compressor and the corresponding operating condition data. Based on the multi-view feature vector, the operating condition data and the underlying causes of vibration, a causal relationship model is constructed between the operating condition variables, vibration feature variables and vibration underlying cause variables. The module distinguishes between vibration features driven by operating condition variables and vibration features driven by vibration underlying causes, and generates a vibration mechanism prior profile based on the causal relationship model.

[0035] The vibration prediction component calculation module is used to collect the current vibration data of the air compressor, decompose the current vibration data into multiple vibration prediction components corresponding to the underlying causes of vibration through the signal decomposition method, and constrain the corresponding vibration prediction components through the vibration mechanism prior profile to obtain the optimized vibration prediction components.

[0036] The fault determination module is used to evaluate the performance of each optimized vibration prediction component in the multi-view feature vector, assign a vibration cause contribution degree to each optimized vibration prediction component, sort the vibration causes based on the vibration cause contribution degree, and form a fault probability judgment.

[0037] Therefore, the present invention provides the following effects and / or advantages:

[0038] This application decomposes the vibration signal of an air compressor into multiple vibration prediction components based on the acceleration, velocity, and displacement perspectives of vibration intensity. A causal relationship model is established to distinguish between vibration characteristics driven by operating condition variables and those driven by underlying vibration causes. A priori profile of the vibration mechanism is generated based on this causal relationship model. Then, the multiple vibration components are optimized according to various underlying vibration causes to obtain more accurate optimized vibration prediction components. The optimized vibration prediction components are ranked by their contribution to the underlying vibration causes to ultimately identify the primary vibration causes.

[0039] This application combines vibration intensity characterization, multi-perspective feature modeling, multi-source data fusion, prior constraints on vibration mechanisms, and contribution assessment based on energy proportion to achieve interpretable, traceable, and rankable early warning of abnormal states in air compressors. Compared to relying solely on a single vibration index for judgment, this embodiment can effectively distinguish the degree of influence of different underlying vibration causes on the current vibration state under complex operating conditions and multiple potential faults, thereby improving the accuracy, stability, and engineering applicability of the early warning system.

[0040] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0041] It should be understood that the above summary and the following detailed description of the invention are exemplary and explanatory, and are intended to provide further explanation of the invention as claimed. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating one embodiment of the present invention. Detailed Implementation

[0043] To facilitate understanding by those skilled in the art, the present invention will now be described in further detail with reference to the embodiments:

[0044] refer to Figure 1 A method for predicting air compressor anomalies includes the following steps:

[0045] S1. Collect historical vibration data of the air compressor under different operating conditions, extract the acceleration perspective, velocity perspective, and displacement perspective of the air compressor from the historical vibration data, calculate the statistical characteristics of the acceleration perspective, velocity perspective, and displacement perspective respectively, and construct a multi-view feature vector of the air compressor vibration.

[0046] In this step, vibration features from the perspectives of acceleration, velocity, and displacement are constructed from historical vibration data. Combined with statistical features, a series of parameters from multiple perspectives can be obtained to describe the vibration intensity of the air compressor. The vibration intensity from different perspectives covers different types of anomalies, providing a stable feature basis for subsequent mechanism learning and decomposition.

[0047] S2, collect the historical underlying causes of vibration of the air compressor and corresponding operating condition data, and construct a causal relationship model between operating condition variables, vibration feature variables and vibration underlying cause variables based on the multi-view feature vector, the operating condition data and the underlying causes of vibration, distinguish the vibration features driven by the operating condition variables and the vibration features driven by the underlying causes of vibration, and generate a priori profile of the vibration mechanism based on the causal relationship model.

[0048] This step can extract operating condition data and corresponding historical vibration underlying causes from historical maintenance work orders. Based on the operating condition data and corresponding historical vibration underlying causes, establish causal relationships between variables to distinguish whether vibration characteristics are driven by operating condition variables or by vibration underlying causes, and generate a priori profile of vibration mechanism for subsequent vibration decomposition constraints.

[0049] S3, collect the current vibration data of the air compressor, decompose the current vibration data into multiple vibration prediction components corresponding to the underlying causes of vibration through the signal decomposition method, constrain the corresponding vibration prediction components through the vibration mechanism prior profile, and obtain the optimized vibration prediction components.

[0050] This step uses signal decomposition to break down the current vibration data into multiple vibration prediction components related to the underlying cause. Combined with prior knowledge of the vibration mechanism, it performs constraint optimization, effectively alleviating the problems of modal aliasing and unclear physical meaning in traditional signal decomposition methods. It suppresses noise components or operational interference components inconsistent with the target vibration mechanism and enhances the frequency band, morphology, and statistical characteristics consistent with the specific underlying cause of vibration. This makes the decomposed vibration prediction components clearer in physical meaning and more stable in statistical characteristics, providing a reliable signal foundation for causal analysis.

[0051] S4. Evaluate the performance of each optimized vibration prediction component in the multi-view feature vector, assign a vibration cause contribution to each optimized vibration prediction component, rank the vibration causes based on the vibration cause contribution, and form a fault probability judgment.

[0052] Based on the optimized vibration prediction components, this step uses the energy proportion of each component within the corresponding frequency band of the current vibration data as the contribution of the vibration cause. The energy proportion directly reflects the actual contribution of each underlying vibration cause to the overall vibration intensity and has a clear physical meaning. This avoids the instability caused by complex model inferences and makes the fault probability judgment process more transparent and interpretable.

[0053] In this embodiment, vibration intensity characterization, multi-perspective feature modeling, multi-source data fusion, prior constraints on vibration mechanisms, and contribution assessment based on energy proportion are combined to achieve interpretable, traceable, and rankable early warning of abnormal states in air compressors. Compared to relying solely on a single vibration index for judgment, this embodiment can effectively distinguish the degree of influence of different underlying vibration causes on the current vibration state under complex operating conditions and multiple potential faults, thereby improving the accuracy, stability, and engineering applicability of the early warning.

[0054] Further, in S1, extracting the acceleration perspective, velocity perspective, and displacement perspective of the air compressor from the historical vibration data includes: the historical vibration data includes an acceleration signal; the acceleration signal is high-pass filtered to obtain the acceleration perspective; the acceleration signal is integrated once and then band-pass filtered at a mid-frequency to obtain the velocity perspective; the acceleration signal is integrated twice and then low-pass filtered to obtain the displacement perspective.

[0055] In this step, a triaxial accelerometer can be installed on the bearing housing of the air compressor to sample acceleration signals. Simultaneously, data cleaning is performed, and the data is sliced ​​according to a fixed duration as the window length to obtain historical vibration data. To prevent drift and other interference in the original signal from entering other perspectives through integration, and to prevent the interference signal from being amplified through integration, this embodiment avoids this by simultaneously operating integration and frequency band control. The high-pass filter can be set to 1000Hz as the cutoff frequency, the mid-frequency bandpass filter can be set to 10-1000Hz as the pass frequency, and the low-pass filter can be set to 10Hz as the cutoff frequency.

[0056] The statistical characteristics include one or more of the following in terms of time and frequency: RMS, peak value, peak-to-peak value, variance, skewness, kurtosis, peak factor, impulse factor, and margin factor.

[0057] In this embodiment, since abnormal vibration of air compressors usually manifests as statistical changes in vibration energy level, impact degree, and waveform structure over a certain time range, rather than anomalies in a single instantaneous value, this method extracts statistical features from vibration signals from the perspectives of acceleration, velocity, and displacement to quantitatively characterize vibration intensity, impact, and waveform distribution features. This provides a stable and physically meaningful feature basis for subsequent vibration cause modeling, constraint decomposition, and anomaly contribution assessment. RMS represents the energy intensity of the vibration signal, which is directly related to mechanical stress, fatigue, and wear. Peak value indicates the presence of spikes or impacts in the waveform, which is directly related to bearing defects, loosening, and rubbing. Peak-to-peak value is highly sensitive to structural loosening and gap-related problems. Peak factor measures the proportion of spikes relative to the overall energy, which is related to impact-related faults. Impulse factor and margin factor are more sensitive to occasional, low-frequency, high-amplitude impacts and are strongly correlated with early bearing spalling, initial loosening, and intermittent rubbing.

[0058] Furthermore, in step S2, by collecting historical maintenance work orders, maintenance behaviors are extracted from the historical maintenance work orders, and the underlying causes of the historical vibrations are inferred based on the maintenance behaviors;

[0059] The operating data includes one or more of the following: air compressor speed, load rate, exhaust pressure, temperature, and current.

[0060] In this step, the historical maintenance work order can record the maintenance behavior of the equipment, such as replacing bearings, aligning couplings, correcting dynamic balance, tightening foundation feet, reinforcing the structure, cleaning foreign objects, adjusting air passages, etc., as well as the corresponding replacement parts, maintenance duration, etc. Through this information, the underlying cause of the historical vibration can be inferred, as shown in Table 1.

[0061] Table 1. Mapping Table of Maintenance Behaviors and Underlying Causes of Vibration

[0062]

[0063] Further, step S2 includes:

[0064] S2.1, construct a benchmark prediction model containing its own hysteresis term for each of the multi-view feature vectors based on the vibration feature variables, and construct an extended prediction model by adding the hysteresis term of the corresponding working condition variable to the benchmark prediction model.

[0065] In this step, the vibration characteristic variables of the multi-view feature vector are divided into fixed time windows. For each time window, the acceleration, velocity, and displacement view features are calculated, and the corresponding time period operating condition variables (speed, load rate, exhaust pressure, etc.) are read synchronously to form a set of time series samples.

[0066] Then, an autoregressive model containing only its own lag terms is constructed for the time series samples corresponding to each vibration characteristic variable to obtain the baseline prediction model. The lag terms of the operating condition variables are added to the baseline prediction model to obtain the extended prediction model.

[0067] S2.2, compare the prediction error difference between the benchmark prediction model and the extended prediction model. When the prediction error difference is greater than a preset threshold, establish the causal relationship edge between the corresponding working condition variable and the vibration characteristic variable and its causal strength value.

[0068] In this step, the prediction error difference can be the mean square error. When the mean square error is greater than a preset threshold, it is considered that the corresponding operating condition variable has a causal driving relationship with the vibration characteristic. The magnitude of the mean square error can be used as the causal strength value.

[0069] S2.3, Perform statistical differentiation analysis between vibration underlying cause variables and vibration characteristic variables under the same working conditions. When the difference in vibration characteristic distribution corresponding to different vibration underlying causes is greater than the preset differentiation threshold, establish the causal relationship edge from the vibration underlying cause variable to the vibration characteristic variable and its causal strength value.

[0070] In this step, to avoid interference from operating conditions, historical samples are grouped according to stable operating conditions. For example, operating conditions with a rotational speed change rate lower than a preset threshold are grouped together. Then, for each vibration characteristic variable, the mean and variance, and other statistical characteristics corresponding to different underlying cause categories of vibration are calculated. When the distribution difference between different cause categories exceeds a preset threshold, a causal relationship edge is established between the underlying cause variable of vibration and that vibration characteristic.

[0071] S2.4 Based on the causal relationship edges and their causal strength values ​​established in steps S2.2 and S2.3, construct a causal relationship model that includes working condition variable nodes, vibration characteristic nodes, and vibration underlying cause nodes;

[0072] In this step, we can traverse all the causal relationship edges and their causal strength values ​​between all the working condition variables and the vibration characteristic variables in steps S2.2 and S2.3, as well as all the causal relationship edges and their causal strength values ​​between all the underlying cause variables of vibration and the vibration characteristic variables, thereby constructing a causal relationship model.

[0073] S2.5, Read the set of parent nodes and their causal intensity values ​​corresponding to each vibration feature node in the causal relationship model, calculate the sum of causal intensity from the working condition variable node for each vibration feature node to obtain the working condition driving intensity, calculate the sum of causal intensity from the vibration underlying cause node for each vibration feature node to obtain the mechanism driving intensity, when the mechanism driving intensity is greater than the working condition driving intensity, determine the vibration feature as a vibration feature driven by the vibration underlying cause, and generate a vibration mechanism prior profile.

[0074] A priori profile of vibration mechanisms is used to impose physical constraints on each underlying cause of vibration, ensuring that it conforms to actual physical characteristics, rather than simply outputting a corresponding waveform. A priori profile of vibration mechanisms can specifically include:

[0075] 1. Statistically analyze the multi-view characteristic distribution range of the same underlying cause of vibration under different working conditions, such as the typical energy concentration frequency band of a certain underlying cause of vibration;

[0076] 2. Extract the typical vibration intensity characteristic combination of the underlying cause of the vibration from the perspectives of acceleration, velocity and displacement, such as the reasonable range of RMS, kurtosis and peak factor of the underlying cause of the vibration under various working conditions.

[0077] 3. Form a set of prior parameters to describe the mechanism characteristics of the vibration cause. The set of prior parameters is used to constrain the subsequent vibration signal decomposition process, such as whether it is an impact type, a harmonic type, or a modulation type.

[0078] Further, in step S3, the current vibration data is decomposed into multiple vibration prediction components corresponding to underlying causes of vibration using a signal decomposition method, including:

[0079] The current vibration data is decomposed into multiple intrinsic mode functions and a residual term using an integrated empirical mode algorithm.

[0080] Calculate the statistical and frequency domain characteristics of each intrinsic mode function, and determine the contribution of each intrinsic mode function to each vibration underlying cause based on the matching relationship between the statistical and frequency domain characteristics and the prior profile of the vibration underlying cause. Determine the vibration underlying cause corresponding to each intrinsic mode function based on the maximum contribution value.

[0081] By weighted reconstructing the intrinsic mode functions of the same underlying cause of vibration, the vibration prediction component of the underlying cause of vibration is obtained.

[0082] In this step, the integrated empirical modal algorithm is a direct adoption of existing technology. By inputting the current vibration data into the algorithm, multiple intrinsic modal functions and a residual term can be obtained. Then, the statistical and frequency domain characteristics of each intrinsic modal function are calculated. For example, the dominant frequency of one intrinsic modal function is approximately 25 Hz (1X revolutions), the RMS is moderate, the kurtosis is approximately 3 (close to normal), and the axial energy proportion is low. The existing a priori profile of the vibration mechanism and its characteristics include: ① Rotor imbalance profile: energy is concentrated at 1X revolutions, kurtosis is low (close to normal), and the axial component is weak; ② Shaft misalignment profile: 1X + 2X is obvious, axial energy is high, and harmonics are abundant; ③ Bearing failure profile: high-frequency impact, significantly large kurtosis, and characteristic frequencies appear in the envelope spectrum. At this point, the characteristic distance matching method can be used to calculate the distance between the intrinsic mode function and the prior profiles of the three vibration mechanisms. The closer the distance, the higher the matching degree. The matching degree between the intrinsic mode function and the rotor imbalance profile is 0.85, the matching degree with the shaft misalignment is 0.30, and the matching degree with the bearing failure is 0.05. Thus, it is determined that the underlying cause of vibration corresponding to the intrinsic mode function is rotor imbalance.

[0083] Furthermore, by combining and reconstructing the intrinsic mode functions of the same underlying cause of vibration, the vibration prediction components of that underlying cause are obtained, including:

[0084] The maximum contribution of each intrinsic mode function is normalized and used as the weight. The intrinsic mode functions of the same underlying cause of vibration are weighted and summed to obtain the vibration prediction component of the underlying cause of vibration.

[0085] Further, in step S3, the corresponding vibration prediction components are constrained by the prior profile of the vibration mechanism to obtain optimized vibration prediction components, including:

[0086] The underlying cause of vibration corresponding to the vibration prediction component is obtained, as well as the prior profile of the vibration mechanism corresponding to the underlying cause of vibration. Based on the prior profile of the vibration mechanism, the vibration prediction component is constrained in terms of frequency band, morphology and statistical characteristics. Components that do not conform to the mechanism are suppressed and components that conform to the mechanism are enhanced. The combined components are the current vibration data, thereby obtaining the optimized vibration prediction component.

[0087] This step determines whether the component is primarily concentrated within the frequency range consistent with the physical mechanism of the equipment or structure based on its frequency band characteristics. Secondly, it examines the signal morphology to check whether its time-domain or time-frequency performance conforms to typical vibration modes. Simultaneously, it combines statistical characteristics, such as energy distribution and amplitude variation patterns, to conduct a consistency assessment. Components that do not conform to the vibration mechanism characteristics are weakened or suppressed; while components highly consistent with the mechanism are retained or enhanced. Through this constraint and adjustment process, an optimized vibration prediction component that better conforms to the physical mechanism, has less noise, and higher reliability is ultimately obtained.

[0088] Further, in step S4, the performance of each optimized vibration prediction component in the multi-view feature vector is evaluated, and the vibration cause contribution is assigned to each optimized vibration prediction component, including:

[0089] Calculate the energy percentage of the optimized vibration prediction component in the corresponding frequency band of the current vibration data, and use the energy percentage as the contribution of the vibration cause.

[0090] In this step, the underlying cause of vibration corresponding to the optimized vibration prediction component corresponds to a certain frequency band. By calculating the energy proportion of the optimized vibration prediction component in that frequency band, interference from irrelevant frequency bands can be avoided. This is used to reflect the relative contribution of each optimized vibration prediction component to the current vibration data, and based on this, a preliminary fault probability judgment is formed.

[0091] Furthermore, an air compressor anomaly prediction system is provided, comprising the following modules:

[0092] The multi-view feature construction module is used to collect historical vibration data of the air compressor under different operating conditions, extract the acceleration view, velocity view, and displacement view of the air compressor from the historical vibration data, calculate the statistical features of the acceleration view, velocity view, and displacement view respectively, and construct the multi-view feature vector of the air compressor vibration.

[0093] The vibration mechanism prior profile generation module is used to collect the historical underlying causes of vibration of the air compressor and the corresponding operating condition data. Based on the multi-view feature vector, the operating condition data and the underlying causes of vibration, a causal relationship model is constructed between the operating condition variables, vibration feature variables and vibration underlying cause variables. The module distinguishes between vibration features driven by operating condition variables and vibration features driven by vibration underlying causes, and generates a vibration mechanism prior profile based on the causal relationship model.

[0094] The vibration prediction component calculation module is used to collect the current vibration data of the air compressor, decompose the current vibration data into multiple vibration prediction components corresponding to the underlying causes of vibration through the signal decomposition method, and constrain the corresponding vibration prediction components through the vibration mechanism prior profile to obtain the optimized vibration prediction components.

[0095] The fault determination module is used to evaluate the performance of each optimized vibration prediction component in the multi-view feature vector, assign a vibration cause contribution degree to each optimized vibration prediction component, sort the vibration causes based on the vibration cause contribution degree, and form a fault probability judgment.

[0096] The principle of this system is the same as that of an air compressor anomaly prediction method, so it will not be described in detail here.

[0097] 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.

[0098] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0099] 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.

[0100] 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.

[0101] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

Claims

1. A method for predicting air compressor anomalies, characterized in that: Includes the following steps: S1. Collect historical vibration data of the air compressor under different operating conditions, extract the acceleration perspective, velocity perspective, and displacement perspective of the air compressor from the historical vibration data, calculate the statistical characteristics of the acceleration perspective, velocity perspective, and displacement perspective respectively, and construct a multi-view feature vector of the air compressor vibration. S2, collect the historical underlying causes of vibration of the air compressor and corresponding operating condition data, and construct a causal relationship model between operating condition variables, vibration feature variables and vibration underlying cause variables based on the multi-view feature vector, the operating condition data and the underlying causes of vibration, distinguish the vibration features driven by the operating condition variables and the vibration features driven by the underlying causes of vibration, and generate a priori profile of the vibration mechanism based on the causal relationship model. Step S2 includes: S2.1, construct a benchmark prediction model containing its own hysteresis term for each of the multi-view feature vectors based on the vibration feature variables, and construct an extended prediction model by adding the hysteresis term of the corresponding working condition variable to the benchmark prediction model. S2.2, compare the prediction error difference between the benchmark prediction model and the extended prediction model. When the prediction error difference is greater than a preset threshold, establish the causal relationship edge between the corresponding working condition variable and the vibration characteristic variable and its causal strength value. S2.3, Perform statistical differentiation analysis between vibration underlying cause variables and vibration characteristic variables under the same working conditions. When the difference in vibration characteristic distribution corresponding to different vibration underlying causes is greater than the preset differentiation threshold, establish the causal relationship edge from the vibration underlying cause variable to the vibration characteristic variable and its causal strength value. S2.4 Based on the causal relationship edges and their causal strength values ​​established in steps S2.2 and S2.3, construct a causal relationship model that includes working condition variable nodes, vibration characteristic nodes, and vibration underlying cause nodes; S2.5, Read the set of parent nodes and their causal intensity values ​​corresponding to each vibration feature node in the causal relationship model, calculate the sum of causal intensity from the working condition variable node for each vibration feature node to obtain the working condition driving intensity, calculate the sum of causal intensity from the vibration underlying cause node for each vibration feature node to obtain the mechanism driving intensity, when the mechanism driving intensity is greater than the working condition driving intensity, determine the vibration feature as a vibration feature driven by the vibration underlying cause, and generate a vibration mechanism prior profile; S3, collect the current vibration data of the air compressor, decompose the current vibration data into multiple vibration prediction components corresponding to the underlying causes of vibration through the signal decomposition method, constrain the corresponding vibration prediction components through the vibration mechanism prior profile, and obtain the optimized vibration prediction components. S4. Evaluate the performance of each optimized vibration prediction component in the multi-view feature vector, assign a vibration cause contribution to each optimized vibration prediction component, rank the vibration causes based on the vibration cause contribution, and form a fault probability judgment.

2. The method for predicting air compressor anomalies according to claim 1, characterized in that: In S1, extracting the acceleration perspective, velocity perspective, and displacement perspective of the air compressor from the historical vibration data includes: the historical vibration data includes an acceleration signal; the acceleration signal is high-pass filtered to obtain the acceleration perspective; the acceleration signal is integrated once and then band-pass filtered at a mid-frequency to obtain the velocity perspective; and the acceleration signal is integrated twice and then low-pass filtered to obtain the displacement perspective. The statistical characteristics include one or more of the following in terms of time and frequency: RMS, peak value, peak-to-peak value, variance, skewness, kurtosis, peak factor, impulse factor, and margin factor.

3. The method for predicting air compressor anomalies according to claim 1, characterized in that: In step S2, by collecting historical maintenance work orders, maintenance behaviors are extracted from the historical maintenance work orders, and the underlying causes of the historical vibrations are inferred based on the maintenance behaviors. The operating data includes one or more of the following: air compressor speed, load rate, exhaust pressure, temperature, and current.

4. The method for predicting air compressor anomalies according to claim 1, characterized in that: In step S3, the current vibration data is decomposed into multiple vibration prediction components corresponding to underlying causes of vibration using a signal decomposition method, including: The current vibration data is decomposed into multiple intrinsic mode functions and a residual term using an integrated empirical mode algorithm. Calculate the statistical and frequency domain characteristics of each intrinsic mode function, and determine the contribution of each intrinsic mode function to each vibration underlying cause based on the matching relationship between the statistical and frequency domain characteristics and the prior profile of the vibration underlying cause. Determine the vibration underlying cause corresponding to each intrinsic mode function based on the maximum contribution value. By combining and reconstructing the intrinsic mode functions of the same underlying cause of vibration, the vibration prediction component of the underlying cause of vibration is obtained.

5. The method for predicting air compressor anomalies according to claim 4, characterized in that: By combining and reconstructing the intrinsic mode functions of the same underlying cause of vibration, the vibration prediction components of that underlying cause are obtained, including: The maximum contribution of each intrinsic mode function is normalized and used as the weight. The intrinsic mode functions of the same underlying cause of vibration are weighted and summed to obtain the vibration prediction component of the underlying cause of vibration.

6. The method for predicting air compressor anomalies according to claim 4, characterized in that: In step S3, the corresponding vibration prediction components are constrained by the prior profile of the vibration mechanism to obtain the optimized vibration prediction components, including: The underlying cause of vibration corresponding to the vibration prediction component is obtained, as well as the prior profile of the vibration mechanism corresponding to the underlying cause of vibration. Based on the prior profile of the vibration mechanism, the vibration prediction component is constrained in terms of frequency band, morphology and statistical characteristics. Components that do not conform to the mechanism are suppressed and components that conform to the mechanism are enhanced. The combined components are the current vibration data, thereby obtaining the optimized vibration prediction component.

7. The method for predicting air compressor anomalies according to claim 1, characterized in that: In step S4, the performance of each optimized vibration prediction component in the multi-view feature vector is evaluated, and the vibration cause contribution is assigned to each optimized vibration prediction component, including: Calculate the energy percentage of the optimized vibration prediction component in the corresponding frequency band of the current vibration data, and use the energy percentage as the contribution of the vibration cause.

8. An air compressor anomaly prediction system, characterized in that: Includes the following modules: The multi-view feature construction module is used to collect historical vibration data of the air compressor under different operating conditions, extract the acceleration view, velocity view, and displacement view of the air compressor from the historical vibration data, calculate the statistical features of the acceleration view, velocity view, and displacement view respectively, and construct the multi-view feature vector of the air compressor vibration. The vibration mechanism prior profile generation module is used to collect historical vibration underlying causes and corresponding operating condition data of the air compressor. Based on the multi-view feature vector, the operating condition data, and the vibration underlying causes, it constructs a causal relationship model between operating condition variables, vibration feature variables, and vibration underlying cause variables. It distinguishes between vibration features driven by operating condition variables and vibration features driven by vibration underlying causes, and generates a vibration mechanism prior profile based on the causal relationship model. This includes: constructing a baseline prediction model containing its own lag term for each of the multi-view feature vectors based on the vibration feature variables; and constructing an extended prediction model by adding the lag term of the corresponding operating condition variable to the baseline prediction model. Compare the prediction error difference between the baseline prediction model and the extended prediction model. When the prediction error difference is greater than a preset threshold, establish a causal relationship edge between the corresponding working condition variable and the vibration characteristic variable, and its causal strength value. A statistical differentiation analysis is performed on the underlying cause variables and vibration characteristic variables under the same working conditions. When the difference in vibration characteristic distribution corresponding to different underlying causes of vibration is greater than the preset differentiation threshold, a causal relationship edge and its causal strength value are established between the underlying cause variables of vibration and the vibration characteristic variables. Based on the established causal relationship edges and their causal strength values, a causal relationship model is constructed that includes working condition variable nodes, vibration characteristic nodes, and vibration underlying cause nodes. Read the set of parent nodes and their causal intensity values ​​corresponding to each vibration feature node in the causal relationship model. Calculate the sum of causal intensity from the working condition variable nodes for each vibration feature node to obtain the working condition driving intensity. Calculate the sum of causal intensity from the underlying vibration cause nodes for each vibration feature node to obtain the mechanism driving intensity. When the mechanism driving intensity is greater than the working condition driving intensity, the vibration feature is determined to be a vibration feature driven by the underlying vibration cause, and a vibration mechanism prior profile is generated. The vibration prediction component calculation module is used to collect the current vibration data of the air compressor, decompose the current vibration data into multiple vibration prediction components corresponding to the underlying causes of vibration through the signal decomposition method, and constrain the corresponding vibration prediction components through the vibration mechanism prior profile to obtain the optimized vibration prediction components. The fault determination module is used to evaluate the performance of each optimized vibration prediction component in the multi-view feature vector, assign a vibration cause contribution degree to each optimized vibration prediction component, sort the vibration causes based on the vibration cause contribution degree, and form a fault probability judgment.

Images