Intelligent analysis method, system and storage medium for fault diagnosis of industrial equipment

Abstract

The present invention provides a fault diagnosis intelligent analysis method, system and storage medium for industrial equipment. The method comprises the following steps: dividing monitoring time series data into at least one subsequence based on the statistical characteristics and working conditions of the monitoring time series data; determining an embedding window length based on the local characteristics of the subsequence, constructing a trajectory matrix using the embedding window length, performing singular value decomposition on the trajectory matrix to obtain a primitive matrix and corresponding singular values; forming candidate groups for primitive matrices of different preset combinations, calculating a separability metric value for each candidate group, and determining component groups based on the separability metric value and the singular value; performing diagonal averaging on the selected component groups to reconstruct signal components, and calculating preset characteristic parameters for each signal component; screening target signal components related to a preset fault mode based on the preset characteristic parameters, extracting characteristic vectors representing the fault state from the target signal components, inputting the characteristic vectors into a preconfigured fault diagnosis model, and outputting fault diagnosis information.

Description

Technical Field

[0001] The present application relates to the field of fault diagnosis, and in particular to a fault diagnosis intelligent analysis method, system and storage medium for industrial equipment. Background Art

[0002] During the long-term service life of industrial equipment, various rotating machinery, reciprocating machinery, and complex equipment systems will inevitably experience various types of failures due to factors such as fatigue, wear, aging, or external impact. If these failures are not diagnosed and addressed promptly and accurately, they can lead to reduced equipment performance and increased energy consumption at best, or catastrophic downtime accidents at worst, causing huge economic losses and even casualties. Traditional equipment maintenance strategies often rely on regular inspections or post-maintenance repairs. The former may lead to excessive or insufficient maintenance, while the latter is often accompanied by high fault repair costs and production interruption losses. Predictive maintenance based on condition monitoring analyzes various monitoring data generated by equipment during operation, such as vibration, current, temperature, pressure, and acoustic signals, to assess the health of the equipment in real time, and can diagnose and warn of failures in the early stages.

[0003] Singular spectrum analysis (SSA) is a nonparametric time series analysis method that, without requiring a precise physical model, adaptively decomposes one-dimensional time series into several additive, physically meaningful signal components. It is particularly well-suited for processing nonlinear and nonstationary industrial vibration signals. Its basic principles include constructing a trajectory matrix, performing singular value decomposition (SVD) on the trajectory matrix, grouping the singular values ​​and their corresponding primitive matrices, and reconstructing the individual components of the original signal through diagonal averaging. However, if the embedding window length is too small, it may not effectively identify long-periodic components, while if the window length is too large, it may smooth out short-term impulse features or introduce excessive noise. Furthermore, the lack of an objective optimization criterion for grouping can easily lead to aliasing of useful signals and noise, compromising the accurate extraction of subsequent fault signatures. Targetedly selecting the components most relevant to a specific fault mode from the numerous signal components decomposed by SSA is crucial for improving its application in industrial equipment fault diagnosis. Summary of the Invention

[0004] To address the technical problem of inaccurate identification caused by inaccurate windows and grouping in existing singular spectrum analysis, this application proposes an intelligent analysis method for fault diagnosis of industrial equipment, including:

[0005] Acquire monitoring time series data of the industrial equipment to be diagnosed, and divide the monitoring time series data into at least one subsequence based on statistical characteristics and operating conditions of the monitoring time series data;

[0006] Determining an embedding window length based on local characteristics of the subsequence, constructing a trajectory matrix using the embedding window length, performing singular value decomposition on the trajectory matrix to obtain a primitive matrix and corresponding singular values; forming candidate groups for primitive matrices of different preset combinations, calculating a separability metric for each candidate group, and determining component groups based on the separability metric and the singular values; performing diagonal averaging on the selected component groups to reconstruct signal components, and calculating preset characteristic parameters for each signal component;

[0007] According to the preset characteristic parameters, target signal components related to the preset fault mode are screened out, feature vectors representing the fault state are extracted from the target signal components, the feature vectors are input into a preconfigured fault diagnosis model, and fault diagnosis information is output.

[0008] Optionally, dividing the monitoring time series data into at least one subsequence based on the statistical characteristics and working conditions of the monitoring time series data includes:

[0009] Calculating statistical properties of the monitoring time series data, including mean, variance, skewness, kurtosis, autocorrelation coefficient and / or power spectral density;

[0010] According to preset operating condition rules, identify operating condition change points in the monitoring time series data, including equipment start and stop, load change, and speed change;

[0011] The monitoring time series data is divided into at least one subsequence according to the statistical characteristics and the operating condition change point.

[0012] Optionally, determining the embedding window length based on the local characteristics of the subsequence includes:

[0013] Calculating local characteristics of the subsequence, wherein the local characteristics include preliminary estimates of data fluctuation amplitude and suspected periodic components;

[0014] The embedding window is preliminarily determined according to the data fluctuation amplitude, and the embedded window is adjusted to obtain the embedded window length L using the suspected periodic component, where the value range of L is [2, N / 2], and N is the length of the subsequence.

[0015] Optionally, calculating the separability metric value of each candidate group includes:

[0016] A separability metric between the signal components in each candidate group is calculated, where the separability metric is a cross-correlation coefficient, a frequency band overlap, or mutual information.

[0017] Optionally, determining the component groups based on the separability measure and the singular value is specifically:

[0018] For each candidate group, the sum of squares of the singular values ​​is calculated to obtain the total energy of the candidate group;

[0019] The total energy and the separability measure are weightedly summed to obtain the scores of the candidate groups, and at least one candidate group with the largest score is used as a component group.

[0020] Optionally, performing diagonal averaging on the selected component groups to reconstruct the signal components and calculating preset characteristic parameters of each signal component includes:

[0021] Perform diagonal averaging reconstruction on the selected component groups to obtain signal components;

[0022] Calculate a preset characteristic parameter of each signal component, wherein the preset characteristic parameter is an energy proportion or an entropy value; wherein the energy proportion is calculated as follows: the energy of the signal component divided by the total energy of all signal components.

[0023] Optionally, screening out target signal components related to a preset fault mode includes:

[0024] According to the characteristics of the preset fault mode, the screening conditions of the target signal components are set;

[0025] Calculate the preset characteristic parameters of each signal component and compare them with the screening conditions;

[0026] The signal components that meet the screening conditions are screened out as target signal components.

[0027] Optionally, extracting a feature vector representing the fault state from the target signal component includes:

[0028] Extracting time domain features from target signal components, wherein the time domain features include mean, variance, skewness, kurtosis, and autocorrelation coefficient;

[0029] Extracting frequency domain features from the target signal component, wherein the frequency domain features include power spectrum density, frequency band energy, and frequency band center frequency;

[0030] The time domain features and the frequency domain features are combined to form a feature vector, where the dimension of the feature vector is the sum of the dimensions of the time domain features and the frequency domain features.

[0031] This application also proposes a fault diagnosis and intelligent analysis system for industrial equipment, comprising:

[0032] an acquisition unit, configured to acquire monitoring time series data of the industrial equipment to be diagnosed, and to divide the monitoring time series data into at least one subsequence based on statistical characteristics and operating conditions of the monitoring time series data;

[0033] a characteristic parameter determination unit, configured to determine an embedding window length based on local characteristics of the subsequence, construct a trajectory matrix using the embedding window length, perform singular value decomposition on the trajectory matrix to obtain a primitive matrix and corresponding singular values; form candidate groups for primitive matrices of different preset combinations, calculate a separability metric value for each candidate group, determine component groups based on the separability metric value and the singular value; reconstruct the selected component groups by diagonal averaging to obtain signal components, and calculate preset characteristic parameters for each signal component;

[0034] The analysis unit is used to screen out target signal components related to the preset fault mode based on the preset characteristic parameters, extract feature vectors representing the fault state from the target signal components, input the feature vectors into a preconfigured fault diagnosis model, and output fault diagnosis information.

[0035] Optionally, dividing the monitoring time series data into at least one subsequence based on the statistical characteristics and working conditions of the monitoring time series data includes:

[0036] Calculating statistical properties of the monitoring time series data, including mean, variance, skewness, kurtosis, autocorrelation coefficient and / or power spectral density;

[0037] According to preset operating condition rules, identify operating condition change points in the monitoring time series data, including equipment start and stop, load change, and speed change;

[0038] The monitoring time series data is divided into at least one subsequence according to the statistical characteristics and the operating condition change point.

[0039] Optionally, determining the embedding window length based on the local characteristics of the subsequence includes:

[0040] Calculating local characteristics of the subsequence, wherein the local characteristics include preliminary estimates of data fluctuation amplitude and suspected periodic components;

[0041] The embedding window is preliminarily determined according to the data fluctuation amplitude, and the embedded window is adjusted to obtain the embedded window length L using the suspected periodic component, where the value range of L is [2, N / 2], and N is the length of the subsequence.

[0042] Optionally, calculating the separability metric value of each candidate group includes:

[0043] A separability metric between the signal components in each candidate group is calculated, where the separability metric is a cross-correlation coefficient, a frequency band overlap, or mutual information.

[0044] Optionally, determining the component groups based on the separability measure and the singular value is specifically:

[0045] For each candidate group, the sum of squares of the singular values ​​is calculated to obtain the total energy of the candidate group;

[0046] The total energy and the separability measure are weightedly summed to obtain the scores of the candidate groups, and at least one candidate group with the largest score is used as a component group.

[0047] Optionally, performing diagonal averaging on the selected component groups to reconstruct the signal components and calculating preset characteristic parameters of each signal component includes:

[0048] Perform diagonal averaging reconstruction on the selected component groups to obtain signal components;

[0049] Calculate a preset characteristic parameter of each signal component, wherein the preset characteristic parameter is an energy proportion or an entropy value; wherein the energy proportion is calculated as follows: the energy of the signal component divided by the total energy of all signal components.

[0050] Optionally, screening out target signal components related to a preset fault mode includes:

[0051] According to the characteristics of the preset fault mode, the screening conditions of the target signal components are set;

[0052] Calculate the preset characteristic parameters of each signal component and compare them with the screening conditions;

[0053] The signal components that meet the screening conditions are screened out as target signal components.

[0054] Optionally, extracting a feature vector representing the fault state from the target signal component includes:

[0055] Extracting time domain features from target signal components, wherein the time domain features include mean, variance, skewness, kurtosis, and autocorrelation coefficient;

[0056] Extracting frequency domain features from the target signal component, wherein the frequency domain features include power spectrum density, frequency band energy, and frequency band center frequency;

[0057] The time domain features and the frequency domain features are combined to form a feature vector, where the dimension of the feature vector is the sum of the dimensions of the time domain features and the frequency domain features.

[0058] The present application also proposes a storage medium, in which at least one program code is stored. The at least one program code is loaded and executed by a processor to implement the fault diagnosis intelligent analysis method as described in any possible implementation manner.

[0059] The present application divides the monitoring time series data according to statistical characteristics, preset working conditions, etc., and adaptively determines the embedding window length based on the local characteristics of each subsequence, which can better adapt to the variability of signal characteristics of industrial equipment in different operating stages or different fault states, and avoids the limitations brought by global fixed parameters. Moreover, by introducing separability metrics and singular values ​​to optimize the grouping process of the primitive matrix, a combination method that can achieve the best discrimination between the reconstructed signal components is selected, thereby improving the purity and accuracy of the useful signal components separated from background noise and other interferences. In addition, after reconstructing the signal components, the present application, based on the preset characteristic parameters such as energy, periodicity, complexity, etc. of each component, and combined with the prior knowledge of the preset fault mode, specifically screens out the target signal components that are most relevant to the potential fault, making the subsequent feature extraction and diagnostic analysis more focused and efficient, and enhancing the detection capability of specific fault modes. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of Example 1;

[0061] Figure 2 is a schematic diagram of the embedded window length;

[0062] Figure 3 is a schematic diagram of sequence data;

[0063] Figure 4 Schematic diagram of the singular values ​​of sequence data;

[0064] Figure 5 Schematic diagram of reconstructing components for sequence data. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0066] The terms "first", "second" and corresponding terminology numbers in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances. This is merely a way of distinguishing when describing objects with the same properties in the embodiments of the present application. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, so that a process, method, system, product or apparatus that includes a series of units is not necessarily limited to those units, but may include other units that are not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0067] In addition, in the description of this application, unless otherwise specified, "plurality" means two or more. The term "and / or" or the character " / " in this application is merely a description of the association relationship between related objects, indicating that three relationships can exist. For example, A and / or B, or A / B, can mean: A exists alone, A and B exist at the same time, or B exists alone.

[0068] A specific embodiment is a method for intelligent analysis of fault diagnosis of industrial equipment, such as Figure 1 Shown, including:

[0069] S1, obtaining monitoring time series data of the industrial equipment to be diagnosed, and dividing the monitoring time series data into at least one subsequence based on the statistical characteristics and working conditions of the monitoring time series data;

[0070] Monitoring time series data comes from various sensors installed in key locations on industrial equipment, such as vibration sensors, temperature sensors, acoustic sensors, current sensors, or pressure sensors. These sensors can collect physical quantities reflecting the equipment's operating status in real time or periodically, converting them into digital time series signals. The acquired data can be single-channel or multi-channel, synchronized signals from multiple sensors. Parameters such as the data sampling frequency and recording duration are appropriately set based on the characteristics of the equipment being diagnosed and the typical frequency range of potential failure modes to ensure that sufficient fault information is captured. For example, bearing failures in high-speed rotating machinery require data collected by high-frequency vibration sensors; for certain slowly developing wear processes, vibration or wear data over longer time spans is of interest.

[0071] The operating state of industrial equipment is not always constant. There are different working conditions such as startup, shutdown, speed change, and load change. The analysis of statistical characteristics includes but is not limited to significant changes in indicators such as the mean, variance, energy, kurtosis, and information entropy of the data segment. For example, Bayesian online change point detection, CUSUM algorithm, or hypothesis testing method based on sliding window is used to identify the time point when the statistical properties in the data stream suddenly change, so as to perform segmentation. Segmentation based on preset condition rules depends on the working conditions obtained from the equipment control system such as the SCADA system, or thresholds related to certain measurable parameters such as speed and load current set by expert knowledge. For example, when the speed signal is monitored to exceed a preset value and stabilize for a period of time, it is considered that the equipment has entered a specific working condition, and the data of the corresponding time period is divided into a subsequence. In one embodiment, an unsupervised clustering algorithm such as K-Means clustering based on the short-time Fourier transform energy spectrum characteristics is used to automatically recognize and segment continuous data segments.

[0072] S2, determining an embedding window length based on local characteristics of the subsequence, constructing a trajectory matrix using the embedding window length, performing singular value decomposition on the trajectory matrix to obtain a primitive matrix and corresponding singular values; forming candidate groups for primitive matrices of different preset combinations, calculating a separability metric for each candidate group, and determining component groups based on the separability metric and the singular values; performing diagonal averaging on the selected component groups to reconstruct signal components, and calculating preset characteristic parameters for each signal component;

[0073] In one embodiment, the embedding window length L is determined by analyzing the autocorrelation function of the subsequence and finding the first significant zero crossing point or peak position, which is often related to the main period or characteristic scale of the signal; or calculating the average mutual information of the subsequence. The one-dimensional time subsequence is mapped into a multi-dimensional trajectory matrix by windowing. Figure 2 A schematic diagram of data sequence and window is shown. Specifically, let the subsequence be , ,..., , after determining the embedding window length L, the i-th column of the trajectory matrix X for ,The dimension of the trajectory matrix is ​​L×K, where K=N-L+1.

[0074] Singular value decomposition SVD transforms the trajectory matrix Decompose into ,in are the singular values ​​in descending order, and are the corresponding left and right singular vectors, is the primitive matrix or basic matrix, and d is the rank of the trajectory matrix. The size of each singular value represents the importance of the corresponding primitive matrix in the original signal. Figure 3 is the original signal, Figure 4 is a singular value.

[0075] The original signal may contain a variety of components of different types and scales, such as long-term trends, periodic fluctuations of different frequencies, noise, etc. After the SVD decomposition of these components, their energy and structural information will be dispersed into different primitive matrices. By trying different combinations of primitive matrices, a signal structure with practical significance can be better reconstructed, and it is helpful to separate aliased components. There are many ways to determine the candidate group, including but not limited to the signal characteristics under certain working conditions being reflected by a specific subset of primitive matrices, and the combination can be preset based on this information; the candidate group can also be formed by random combination. In another embodiment, a preliminary division is performed based on the inflection point of the singular value spectrum, or several adjacent primitive matrices are combined. In one embodiment, the separability measure uses the similarity between different pairs of reconstructed components, Figure 5 Different reconstructed components are shown. A similarity value close to 1 indicates that the two components are highly correlated and grouped together, while a similarity value close to 0 indicates that they are uncorrelated and belong to different groups. In another embodiment, distance correlation or a kernel-based method is used. When determining component groups from multiple candidate groups, a combination of separability metrics and singular values ​​is used. Alternatively, component groups can be determined based solely on separability metrics, and the group with the optimal metric is selected so that the components within the group are highly correlated, while the components between the groups are as independent or distinguishable as possible.

[0076] The primitive matrices within the selected group are summed, and then the reconstructed matrices are converted back into one-dimensional time series signal components through a diagonal averaging process, and characteristic parameters of each reconstructed signal component are calculated. In one embodiment, the characteristic parameters include energy parameters, periodicity parameters, and / or complexity parameters. Energy parameters include the root mean square value and variance of the signal, periodicity parameters include extracting the main frequency, harmonic components, spectral peak amplitude, etc. by performing a fast Fourier transform on the signal components, and complexity parameters include permutation entropy, sample entropy, etc.

[0077] S3, based on the preset characteristic parameters, screen out target signal components related to the preset fault mode, extract feature vectors representing the fault state from the target signal components, input the feature vectors into a preconfigured fault diagnosis model, and output fault diagnosis information.

[0078] The screening method includes but is not limited to the characteristic description of a specific fault mode based on the expert knowledge base. For example, if it is known that a certain fault will produce impact vibration within a specific frequency range, the screening condition is set as follows: the energy of the signal component is concentrated in this frequency range, and its kurtosis value or complexity index is significantly higher than that of the normal signal. In another embodiment, a rule-based reasoning system is used for screening, wherein the rule base contains the association between different fault modes and the characteristic parameters of the signal components; or a threshold method is used for screening, for example, the energy proportion of the target signal component exceeds a certain threshold, and its main frequency falls near the preset fault characteristic frequency. After screening out the target signal component, a set of numerical features that characterize the fault state of the equipment is extracted to form a feature vector. The feature vector is input into the fault diagnosis model to obtain fault diagnosis information, and the fault diagnosis model is a support vector machine, a neural network, etc.

[0079] This implementation segments monitoring data based on statistical characteristics and operating conditions. By determining the embedding window length, constructing a trajectory matrix, performing singular value decomposition, and grouping and reconstructing components, signal components are extracted and their characteristic parameters are calculated. Target signal components associated with the pre-defined fault are screened, and fault feature vectors are extracted. Finally, these vectors are input into a diagnostic model to output the results. Data segmentation adapts to the non-stationary nature of equipment operation, while signal decomposition is used to extract fault-related information and generate features, thereby improving the targetedness and accuracy of diagnosis.

[0080] There are often significant differences in the monitoring data of industrial equipment under different operating conditions, such as start-stop, different loads, and different speeds. If the entire time series is analyzed indiscriminately, the data characteristics of different operating conditions will interfere with each other, which may mask the weak characteristics of early faults or cause normal operating condition fluctuations to be misjudged as faults. By segmentation, subsequences with similar characteristics can be separated, so that subsequent feature extraction and fault diagnosis models can more accurately model and analyze data behaviors under specific operating conditions. In an optional embodiment, the monitoring time series data is segmented into at least one subsequence based on the statistical characteristics and operating conditions of the monitoring time series data, including:

[0081] Calculating statistical properties of the monitoring time series data, including mean, variance, skewness, kurtosis, autocorrelation coefficient and / or power spectral density;

[0082] According to preset operating condition rules, identify operating condition change points in the monitoring time series data, including equipment start and stop, load change, and speed change;

[0083] The monitoring time series data is divided into at least one subsequence according to the statistical characteristics and the operating condition change point.

[0084] Specifically, for each data segment within a window, key statistical characteristics are calculated, such as mean, variance / standard deviation, skewness, kurtosis, autocorrelation coefficient, and power spectral density. Furthermore, SCADA, DCS, and PLC systems record timestamps for equipment start / stop commands, load setpoint changes, speed setpoint changes, and other operations. These timestamps directly correspond to operating condition change points. The time points when equipment startup completes and shutdown begins, the start and end times of switching from one stable level to another, and the start and end times of speed adjustment and reaching the new stable speed are recorded. A change in operating conditions signifies a fundamental shift in the equipment's operating mechanism. These identified operating condition change points serve as mandatory split points. Within the data segments initially segmented by these operating condition change points, a change point detection algorithm is then applied to the time series of statistical characteristics, such as mean and variance. If a significant and sustained sudden change in a statistical characteristic is detected, a split point is added at that point, even if there are no externally recorded operating condition changes. A split point is defined when the variance calculated within the sliding window increases by more than a preset percentage within a short period of time.

[0085] The embedding window length is key to constructing the trajectory matrix and directly affects the quality and interpretability of the decomposition results. If the window length is too short, it may not fully capture features in the signal with long periods or slow evolution, such as weak signals of certain early faults or low-frequency vibration modes of equipment operation, resulting in information loss. Conversely, if the window length is too long, it may smooth out important details in the signal or confuse signal components from different physical sources, reducing the resolution of the components and the accuracy of subsequent analysis. In an optional embodiment, the embedding window length is determined based on the local characteristics of the subsequence, including:

[0086] Calculating local characteristics of the subsequence, wherein the local characteristics include preliminary estimates of data fluctuation amplitude and suspected periodic components;

[0087] The embedding window is preliminarily determined according to the data fluctuation amplitude, and the embedded window is adjusted to obtain the embedded window length L using the suspected periodic component, where the value range of L is [2, N / 2], and N is the length of the subsequence.

[0088] In order to evaluate the amplitude of data fluctuations, a sliding window technique is used to calculate statistical indicators in local areas on the subsequence, such as variance, standard deviation, peak-to-peak value or energy. For example, a smaller detection window is set, which slides on the subsequence to calculate the standard deviation sequence of the data in each small window, so as to understand the temporal distribution of signal energy and the degree of change. For a preliminary estimate of suspected periodic components, preferably, the autocorrelation function ACF of the subsequence is calculated to see if there are significant peaks other than zero delay in the ACF. The delay times corresponding to these peaks may indicate the potential period length. In yet another embodiment, the fast Fourier transform of the subsequence is calculated to obtain its power spectral density PSD, and the frequency peaks with significant energy concentration are searched on the power spectrum. The inverses of these frequency peaks also correspond to potential periods.

[0089] In the initial stage, a candidate embedding window range is preliminarily set based on the approximate data fluctuation amplitude. For example, if analysis indicates that the subsequence contains a large number of high-frequency, rapidly changing components, a relatively small initial window length candidate value is preferred to more precisely capture these dynamics. Conversely, if the subsequence exhibits gentle fluctuations and is dominated by low-frequency components, a larger initial window length candidate value is selected. This preliminary embedding window is then adjusted using the estimated suspected periodic components. If one or more significant suspected periods are identified from the ACF or PSD, the embedding window length is set to an integer multiple of these periods, or at least to ensure that L covers the length of the most significant or longest period, for example, L is equal to a certain longest period or twice the longest period. Significance of the suspected periods is determined by their energy, for example, if their energy is greater than the average. This ensures that the periodic components are fully represented within the embedding window and can be effectively separated in the subsequent decomposition. The final embedding window length must also satisfy the constraint that its value range is [2, N / 2], where N is the length of the subsequence being processed. In another embodiment, the adjustment method is to determine an integer multiple of the suspected period to form a set, and adjust the initial window length candidate value to the value closest to the value in the set. For example, if the suspected period is 2 and the set is [2, 4, 8], if the initial window is 3.6, it jumps to 4.

[0090] In order to ensure that each signal component ultimately used for fault diagnosis has a high fault indication significance and minimizes information redundancy, in an optional embodiment, the calculation of the separability metric value of each candidate group includes:

[0091] A separability metric between the signal components in each candidate group is calculated, where the separability metric is a cross-correlation coefficient, a frequency band overlap, or mutual information.

[0092] After performing singular value decomposition (SVD) on the subsequence trajectory matrix, a series of primitive matrices are obtained, sorted by the magnitude of their corresponding singular values. Each primitive matrix consists of a singular value, a corresponding left singular vector, and a corresponding right singular vector. These primitive matrices represent the energy and structure of the signal in different directions. Based on the magnitude of the singular values, primitive matrices with large and similar singular values ​​are grouped together; together, they constitute a primary signal component. Alternatively, primitive matrices with a single, prominent singular value can be considered as a separate component. Another approach is to try grouping the first k (k for different values) primitive matrices into one group, and the remaining into another or multiple groups. Alternatively, primitive matrices with similar characteristics can be grouped together based on a preliminary assessment of the possible signal component types, such as trend terms, periodic terms of specific frequencies, and noise. Each candidate group represents a potential decomposition of the original signal, where each subgroup reconstructs a specific signal component through diagonal averaging.

[0093] For each previously generated candidate group, the independence between different signal components within it is quantitatively evaluated. Specifically, by calculating their Pearson correlation coefficient in the time domain, the closer the value is to 0, the weaker the linear correlation between the two components in time, that is, the more separated they are; or the power spectral density (PSD) of each signal component is calculated, and then, for any two components, the degree of overlap of their power spectra on the frequency axis is analyzed. The smaller the overlap, the more obvious the components are in the frequency domain; mutual information can also be used. The smaller the mutual information value, the lower the statistical dependence between the two components, that is, the less helpful the knowledge of one component is in predicting the other component, and the more separated they are.

[0094] Relying solely on the separability metric may select some components that are statistically independent but have weak physical significance or extremely low energy proportions, while the singular value directly reflects the energy contribution or importance of each signal component in the original signal. By combining the two, component groups that can be clearly separated in signal structure and actually carry the main information of the original signal are selected, thereby improving the effectiveness of subsequent fault feature extraction and the accuracy of diagnostic results. In an optional embodiment, the component group is determined based on the separability metric and the singular value, specifically:

[0095] For each candidate group, the sum of squares of the singular values ​​is calculated to obtain the total energy of the candidate group;

[0096] The total energy and the separability measure are weightedly summed to obtain the scores of the candidate groups, and at least one candidate group with the largest score is used as a component group.

[0097] For each candidate group, its total energy is calculated. The energy is obtained by summing the squares of the singular values ​​corresponding to the primitive matrices in the group. In order to eliminate the dimensional effect and make it comparable, normalization is performed, for example, its proportion in the total energy of all singular values ​​is calculated. It should be noted that some of the operations that may be involved in this application need to be normalized. Those skilled in the art know when normalization should be performed. Conventional operations such as normalization will not be described in detail, but the absence of explanation does not mean that there are no conventional operations such as normalization.

[0098] For each candidate group, a pre-calculated separability metric is used to assess the degree of independence or distinction between signal components within the group. For example, this metric is based on metrics such as the average cross-correlation coefficient, frequency band non-overlap, or mutual information. Larger values ​​indicate better separability. To comprehensively consider both energy contribution and separability, a composite score is calculated for each candidate group. This is achieved by taking a weighted sum of the normalized energy contribution and the separability metric. The scores are sorted from high to low, and the candidate group or groups with the highest scores are selected as the final component group.

[0099] In an optional embodiment, performing diagonal averaging on the selected component groups to reconstruct the signal components and calculating the preset characteristic parameters of each signal component includes:

[0100] Perform diagonal averaging reconstruction on the selected component groups to obtain signal components;

[0101] Calculate a preset characteristic parameter of each signal component, wherein the preset characteristic parameter is an energy proportion or an entropy value; wherein the energy proportion is calculated as follows: the energy of the signal component divided by the total energy of all signal components.

[0102] A selected component group contains one or more primitive matrices. If a component group contains multiple primitive matrices, these primitive matrices are first summed to form a new matrix. The elements on the anti-diagonal of this new matrix are averaged to recover the one-dimensional time series. Preset characteristic parameters, such as energy percentage or entropy value, are calculated for each reconstructed signal component. After calculating these preset characteristic parameters, each signal component is represented by a set of numerical values, such as its energy percentage or a certain entropy value. These values ​​form part of the feature vector subsequently input into the fault diagnosis model.

[0103] After decomposition, the monitoring signal of industrial equipment may produce multiple signal components. These components may correspond to the normal operating status of the equipment, different types of potential faults, background noise or other environmental influences. Not all components are equally valuable for the diagnosis of specific faults. By pre-defining the characteristic parameter range or pattern related to the known fault mode and applying it to the calculated characteristic parameters of each signal component, target-oriented component selection is achieved to ensure that only those components that are most likely to carry the target fault information are further processed, thereby avoiding the waste of computing resources on a large amount of irrelevant data and reducing the risk of normal fluctuations or non-target faults confusing the diagnostic results. In an optional embodiment, the screening out of target signal components related to the preset fault mode includes:

[0104] According to the characteristics of the preset fault mode, the screening conditions of the target signal components are set;

[0105] Calculate the preset characteristic parameters of each signal component and compare them with the screening conditions;

[0106] The signal components that meet the screening conditions are screened out as target signal components.

[0107] Each predefined failure mode or category is determined based on prior knowledge of specific industrial equipment and its common failure modes. A certain failure might significantly increase signal energy within a specific frequency range, making the energy fraction or energy within a specific frequency band a key characteristic parameter. Another failure might reduce signal regularity and increase signal complexity, making sample entropy or permutation entropy more sensitive.

[0108] The screening condition is based on the comparison of thresholds, for example, the energy proportion of a signal component is greater than , or its sample entropy is less than 0.1. It can also be based on pattern matching, for example, the power spectrum of a signal component has an expected peak pattern in a specific frequency band, or it can be based on the combination logic of multiple characteristic parameters, for example, the energy proportion is high AND the energy of a specific frequency band is also high.

[0109] The preset characteristic parameters of each signal component obtained in the previous step are calculated and compared one by one with the screening criteria set in the previous step. Signal components that meet all specified criteria are ultimately selected as target signal components. After decomposing and reconstructing the original subsequence and calculating the preset characteristic parameters such as energy contribution and entropy value for each signal component, these calculated characteristic parameter values ​​are substituted into the screening criteria set for a specific fault mode for evaluation. For each signal component, its calculated characteristic parameters are checked to see if they meet the screening criteria for the preset fault mode A. If so, the component is marked as a target signal component associated with fault mode A. The same process is repeated for preset fault modes B, C, and so on. A signal component may meet the screening criteria for multiple fault modes simultaneously or may not meet the criteria for any preset fault mode. A set of one or more target signal components is output. These are signal portions that are considered highly correlated with the preset fault mode or modes of interest. If no component meets the screening criteria, the current data does not exhibit the preset fault characteristics.

[0110] The original target signal component itself is still time series data. Directly inputting it into the diagnostic model may lead to problems such as excessive dimensionality, information redundancy, and insensitivity to small changes. By extracting time domain and frequency domain features such as mean, variance, power spectrum density, and frequency band energy, the essential information related to the fault state in the signal can be concentrated into a few eigenvalues, which not only greatly reduces the data dimension but also improves the ability to distinguish. In an optional embodiment, the extraction of a feature vector representing the fault state from the target signal component includes:

[0111] Extracting time domain features from target signal components, wherein the time domain features include mean, variance, skewness, kurtosis, and autocorrelation coefficient;

[0112] Extracting frequency domain features from the target signal component, wherein the frequency domain features include power spectrum density, frequency band energy, and frequency band center frequency;

[0113] The time domain features and the frequency domain features are combined to form a feature vector, where the dimension of the feature vector is the sum of the dimensions of the time domain features and the frequency domain features.

[0114] After calculating the time domain features and frequency domain features, they are arranged and combined in a predetermined order to obtain a feature vector, which is then provided as input to subsequent fault diagnosis models such as support vector machines, neural networks, decision trees, etc. for classification.

[0115] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some feature data can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0116] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0117] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0118] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

Claims

1. A fault diagnosis intelligent analysis method for industrial equipment, characterized in that: include: Acquiring monitoring time series data of the industrial equipment to be diagnosed, and dividing the monitoring time series data into at least one subsequence based on statistical characteristics and operating conditions of the monitoring time series data; the monitoring time series data is derived from sensors installed at key locations of the industrial equipment, the sensors including vibration sensors, temperature sensors, acoustic sensors, current sensors, or pressure sensors; and the operating conditions including startup, shutdown, speed change, and load change; Determining an embedding window length based on local characteristics of the subsequence, constructing a trajectory matrix using the embedding window length, and performing singular value decomposition on the trajectory matrix to obtain a primitive matrix and corresponding singular values; For different preset combinations of primitive matrices, candidate groups are formed, separability metrics are calculated for each candidate group, and component groups are determined based on the separability metrics and singular values; signal components are reconstructed by diagonal averaging of the selected component groups, and preset characteristic parameters of each signal component are calculated; According to the preset characteristic parameters, target signal components related to the preset fault mode are screened out, feature vectors representing the fault state are extracted from the target signal components, the feature vectors are input into a preconfigured fault diagnosis model, and fault diagnosis information is output; Calculating the separability metric value of each candidate group includes: Calculating a separability metric between signal components in each candidate group, wherein the separability metric is a cross-correlation coefficient, a frequency band overlap, or a mutual information; The component groups are determined based on the separability measure and the singular value, specifically: For each candidate group, the sum of squares of the singular values ​​is calculated to obtain the total energy of the candidate group; The total energy and the separability measure are weightedly summed to obtain the scores of the candidate groups, and at least one candidate group with the largest score is used as a component group.

2. The method according to claim 1, characterized in that The step of dividing the monitoring time series data into at least one subsequence based on the statistical characteristics and working conditions of the monitoring time series data comprises: Calculating statistical properties of the monitoring time series data, including mean, variance, skewness, kurtosis, autocorrelation coefficient and / or power spectral density; According to preset operating condition rules, identify operating condition change points in the monitoring time series data, including equipment start and stop, load change, and speed change; The monitoring time series data is divided into at least one subsequence according to the statistical characteristics and the operating condition change point.

3. The method according to claim 1, characterized in that The determining of the embedding window length based on the local characteristics of the subsequence includes: Calculating local characteristics of the subsequence, wherein the local characteristics include preliminary estimates of data fluctuation amplitude and suspected periodic components; The embedding window is preliminarily determined according to the data fluctuation amplitude, and the embedded window is adjusted to obtain the embedded window length L using the suspected periodic component, where the value range of L is [2, N / 2], and N is the length of the subsequence.

4. The method according to claim 1, wherein The diagonal averaging and reconstruction of the selected component groups to obtain signal components and calculating the preset characteristic parameters of each signal component include: Perform diagonal averaging reconstruction on the selected component groups to obtain signal components; Calculate a preset characteristic parameter of each signal component, wherein the preset characteristic parameter is an energy proportion or an entropy value; wherein the energy proportion is calculated as follows: the energy of the signal component divided by the total energy of all signal components.

5. The method according to claim 1, wherein The step of screening out target signal components related to a preset fault mode includes: According to the characteristics of the preset fault mode, the screening conditions of the target signal components are set; Calculate the preset characteristic parameters of each signal component and compare them with the screening conditions; The signal components that meet the screening conditions are screened out as target signal components.

6. The method according to claim 1, characterized in that The step of extracting a feature vector representing the fault state from the target signal component includes: Extracting time domain features from target signal components, wherein the time domain features include mean, variance, skewness, kurtosis, and autocorrelation coefficient; Extracting frequency domain features from the target signal component, wherein the frequency domain features include power spectrum density, frequency band energy, and frequency band center frequency; The time domain features and the frequency domain features are combined to form a feature vector, where the dimension of the feature vector is the sum of the dimensions of the time domain features and the frequency domain features.

7. A fault diagnosis intelligent analysis system for industrial equipment, characterized in that: include: An acquisition unit is configured to acquire monitoring time series data of the industrial equipment to be diagnosed, and to segment the monitoring time series data into at least one subsequence based on statistical characteristics and operating conditions of the monitoring time series data; the monitoring time series data is derived from sensors installed at key locations of the industrial equipment, the sensors including vibration sensors, temperature sensors, acoustic sensors, current sensors, or pressure sensors; and the operating conditions including startup, shutdown, speed change, and load change; a characteristic parameter determination unit, configured to determine an embedding window length based on local characteristics of the subsequence, construct a trajectory matrix using the embedding window length, and perform singular value decomposition on the trajectory matrix to obtain a primitive matrix and corresponding singular values; For different preset combinations of primitive matrices, candidate groups are formed, separability metrics are calculated for each candidate group, and component groups are determined based on the separability metrics and singular values; signal components are reconstructed by diagonal averaging of the selected component groups, and preset characteristic parameters of each signal component are calculated; an analysis unit, configured to screen target signal components related to a preset fault mode based on the preset characteristic parameters, extract a feature vector representing the fault state from the target signal components, input the feature vector into a preconfigured fault diagnosis model, and output fault diagnosis information; Calculating the separability metric value of each candidate group includes: Calculating a separability metric between signal components in each candidate group, wherein the separability metric is a cross-correlation coefficient, a frequency band overlap, or a mutual information; The component groups are determined based on the separability measure and the singular value, specifically: For each candidate group, the sum of squares of the singular values ​​is calculated to obtain the total energy of the candidate group; The total energy and the separability measure are weightedly summed to obtain the scores of the candidate groups, and at least one candidate group with the largest score is used as a component group.

8. A storage medium, characterized in that: At least one program code is stored in the storage medium, and the at least one program code is loaded and executed by the processor to implement the fault diagnosis intelligent analysis method according to any one of claims 1 to 6.

Images