Fault prediction method for vulnerable parts in logistics equipment
Through real-time monitoring of vulnerable components of logistics equipment and multi-angle feature analysis, combined with digital twin models and motion analysis, the joint fault prediction results are generated, which solves the problem of inaccurate prediction in the existing technology, and improves the comprehensiveness of fault prediction and Accuracy enables accurate grasp and timely early warning of the status of vulnerable components.
Patent Information
- Application Number
- CN202411063272.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-12-06
AI Technical Summary
The existing methods for predicting defective components of logistics equipment rely on regular maintenance and empirical judgment, lack scientificity and accuracy, especially in complex environments, which makes it difficult to fully reflect the status of the components, resulting in poor prediction results.
Through real-time state monitoring of vulnerable components of logistics equipment, multi-angle feature extraction and fusion, including time-domain and frequency-domain feature analysis, combined with digital twin models and Levi and Brownian motion analysis on multi-time/frequency scales, a joint is generated Failure prediction results.
It improves the comprehensiveness and accuracy of fault prediction, reduces the error of single feature prediction, realizes accurate grasp and timely warning of the state of vulnerable components, and reduces downtime and maintenance costs.
Smart Images

Figure CN119089168A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of intelligent manufacturing technology, and in particular to a method for predicting failures of vulnerable parts in logistics equipment. Background Art
[0002] In the modern logistics industry, the operating efficiency and stability of logistics equipment are directly related to the smooth progress of logistics operations and cost control. The vulnerable parts in logistics equipment, such as transmission gears, bearings, motors, etc., are prone to wear, breakage or performance degradation due to long-term high-load operation and environmental factors. If these faults are not discovered and handled in time, they will seriously affect the efficiency of logistics operations and even paralyze the entire logistics system, causing huge economic losses to the company.
[0003] At present, the traditional method for predicting the failure of vulnerable parts of logistics equipment mainly relies on regular manual inspection and preventive maintenance based on experience judgment. However, this method has many shortcomings: on the one hand, the period of regular maintenance is often set based on experience and lacks scientific basis, which may lead to excessive maintenance or untimely maintenance; on the other hand, manual judgment is limited by the technical level and experience of personnel, and it is difficult to accurately predict all potential failures, especially those small changes that are not easy to detect in the early stage.
[0004] In addition, with the development of the Internet of Things, big data and artificial intelligence technologies, although some data-driven fault prediction methods have been proposed, these methods often rely on only a single type of data feature for prediction, ignoring the complementarity and correlation between different feature parameters, resulting in the accuracy and reliability of the prediction results needing to be improved. Especially in a complex and changeable logistics environment, the prediction model of a single feature parameter is often difficult to fully reflect the actual status of vulnerable parts, thus limiting the effect of fault prediction. Summary of the invention
[0005] The purpose of this application is to propose a fault prediction method for vulnerable parts in logistics equipment, which is used to solve or overcome the above-mentioned technical problems existing in the prior art.
[0006] According to a first aspect of an embodiment of the present application, a method for predicting failure of vulnerable parts in logistics equipment is provided, which includes:
[0007] Performing real-time monitoring on the operating status of the vulnerable parts to obtain real-time status monitoring data of the vulnerable parts;
[0008] Performing a first feature extraction on the real-time status monitoring data to obtain a first feature parameter;
[0009] Performing a second feature extraction on the real-time status monitoring data to obtain a second feature parameter;
[0010] Performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result;
[0011] Performing fault prediction based on the second characteristic parameter to obtain a second fault prediction result;
[0012] fusing the first fault prediction result and the second fault prediction result to obtain a joint fault prediction result for the vulnerable component;
[0013] According to the combined fault prediction result, fault prediction is performed on the vulnerable component.
[0014] Optionally, the real-time monitoring of the operating status of the vulnerable component to obtain real-time status monitoring data of the vulnerable component includes:
[0015] Based on the sensor arranged on the vulnerable part, the operating state of the vulnerable part is monitored in real time to obtain equipment operating state data, and based on the monitored time series, a time series-based equipment operating state data set is formed;
[0016] Preprocessing the equipment operation status data set to obtain a fault prediction data set, wherein the preprocessing includes at least one of cleaning, denoising, and normalization;
[0017] A feature analysis is performed on the fault prediction data set to obtain real-time status monitoring data of the vulnerable parts.
[0018] Optionally, the performing feature analysis on the fault prediction data set to obtain real-time status monitoring data of the vulnerable component therefrom includes:
[0019] A correlation / redundancy analysis is performed on the fault prediction data set to obtain real-time status monitoring data of the vulnerable component therefrom.
[0020] Optionally, the correlation analysis is used to analyze the correlation between the fault prediction data set and the state of the wearing parts to obtain a correlation measurement value, so as to select the equipment operation status data with a correlation measurement value greater than a set correlation threshold as the alternative status monitoring data to form an alternative status monitoring data set; the redundancy analysis is used to analyze the semantic similarity of the alternative status monitoring data in the alternative status monitoring data set to obtain a similarity measurement value, so as to deduplicate the alternative status monitoring data with a similarity measurement value greater than a set similarity threshold to obtain the real-time status monitoring data.
[0021] Optionally, performing a first feature extraction on the real-time status monitoring data to obtain a first feature parameter includes:
[0022] In the time domain, time domain feature extraction is performed on the real-time status monitoring data to obtain time domain feature parameters.
[0023] Optionally, in the time domain, extracting time domain features from the real-time status monitoring data to obtain time domain feature parameters includes:
[0024] Based on the monitored time series, the real-time state monitoring data is sampled in the order of the real-time state monitoring data to obtain real-time state time domain sampling data to form a real-time state time domain sampling data sequence;
[0025] Based on the set feature extraction index, feature extraction is performed on the real-time state time domain sampling data sequence to obtain time domain feature parameters.
[0026] Optionally, the extracting features of the real-time state time domain sampling data sequence based on the set feature extraction index to obtain time domain feature parameters includes:
[0027] Based on the set dimensional index, the semantic feature extraction of the actual physical meaning of the real-time state time domain sampling data sequence is performed to obtain a first time domain feature parameter;
[0028] Based on the set dimensionless index, the real-time state time domain sampling data sequence is distributedly sampled to obtain a second time domain characteristic parameter.
[0029] Optionally, the first time domain characteristic parameter includes: at least one of: mean value, variance, peak-to-peak value, root mean square value, root amplitude, average amplitude, and peak value; the second time domain characteristic parameter includes: at least one of: slope index, kurtosis index, peak factor, margin index, waveform factor, and pulse index.
[0030] Optionally, performing a second feature extraction on the real-time status monitoring data to obtain a second feature parameter includes:
[0031] Based on the time-frequency transformation model, the real-time status monitoring data is transformed from the time domain to the frequency domain to obtain frequency domain status monitoring data, and frequency domain features are extracted from the frequency domain status monitoring data to obtain frequency domain feature parameters.
[0032] Optionally, the time-frequency transformation model is a time-frequency transformation model, a Fourier transformation model and / or a Fourier series decomposition model.
[0033] Optionally, the extracting frequency domain features from the frequency domain state monitoring data to obtain frequency domain feature parameters includes:
[0034] Based on the set periodicity and boundedness constraints, the frequency data of the frequency domain state monitoring data is determined, and frequency domain features of the frequency domain state monitoring data are extracted based on the frequency data, and a correlation relationship between the extracted frequency domain features and the frequency data is established to obtain the frequency domain feature parameters.
[0035] Optionally, establishing an association relationship between the extracted frequency domain features and the frequency data to obtain the frequency domain feature parameters includes:
[0036] Based on the mapping model, the frequency domain characteristics and frequency data are fitted with a correlation relationship to obtain the frequency domain characteristic parameters.
[0037] Optionally, the frequency domain characteristic parameter includes: at least one of: centroid frequency, mean square frequency, root mean square frequency, mean square deviation frequency, and standard deviation frequency.
[0038] Optionally, the mapping model includes at least one of a power spectrum model, a phase spectrum model, an amplitude spectrum model, and an envelope spectrum model.
[0039] Optionally, the performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result includes: predicting an operating state of the vulnerable component based on the first characteristic parameter to obtain a first predicted operating state, and diagnosing a fault of the vulnerable component based on the first predicted operating state to obtain a first fault prediction result;
[0040] The performing fault prediction based on the second characteristic parameter to obtain a second fault prediction result includes: predicting the operating state of the wearing part based on the second characteristic parameter to obtain a second predicted operating state, and diagnosing the fault of the wearing part based on the second predicted operating state to obtain a second fault prediction result.
[0041] Optionally, predicting the operating state of the consumable part based on the first characteristic parameter to obtain a first predicted operating state includes: loading the first characteristic parameter onto a digital twin model of the consumable part to perform a simulation operation to predict the operating state of the consumable part to obtain the first predicted operating state.
[0042] The predicting the operating state of the consumable component based on the second characteristic parameter to obtain a second predicted operating state includes: loading the second characteristic parameter onto a digital twin model of the consumable component to perform a simulation operation to predict the operating state of the consumable component to obtain a second predicted operating state.
[0043] Optionally, performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result includes:
[0044] The first characteristic parameter is input into the constructed fault prediction model to perform fault prediction and obtain a first fault prediction result.
[0045] Optionally, inputting the first characteristic parameter into a constructed fault prediction model to perform fault prediction to obtain a first fault prediction result includes:
[0046] Inputting the first characteristic parameter into the constructed fault prediction model to perform multi-time scale Levy motion and Brownian motion analysis based on the first characteristic parameter to obtain a Levy motion analysis time series and a Brownian motion analysis time series;
[0047] Based on the Levy motion analysis time series and the Brownian motion analysis time series, fault prediction is performed on the vulnerable component to obtain a first fault prediction result.
[0048] Optionally, the step of inputting the first characteristic parameter into a constructed fault prediction model to perform multi-time-scale Levy motion and Brownian motion analysis based on the first characteristic parameter to obtain a Levy motion analysis time series and a Brownian motion analysis time series comprises:
[0049] Based on the constructed fault prediction model, long-tail distribution and self-similarity of the first characteristic parameter at multiple time scales are statistically analyzed to perform Levy motion analysis at multiple time scales on the first characteristic parameter to obtain a Levy motion analysis time series;
[0050] Based on the constructed fault prediction model, the random motion of the first characteristic parameter on multiple time scales is tracked to perform Brownian motion analysis on the first characteristic parameter to obtain a Brownian motion analysis time series.
[0051] Optionally, the performing fault prediction on the vulnerable component based on the Levy motion analysis time series and the Brownian motion analysis time series to obtain a first fault prediction result includes:
[0052] Vectorizing the Levy motion analysis time series and the Brownian motion analysis time series respectively to obtain a Levy motion feature vector and a Brownian motion feature vector;
[0053] Performing vector fusion on the Levy motion feature vector and the Brownian motion feature vector to obtain a fault prediction feature vector;
[0054] The constructed fault prediction model performs walking iteration on the initial fault prediction parameters constructed for the vulnerable component based on the fault prediction feature vector, so as to calculate the time domain development behavior fitness of the vulnerable component to obtain a time domain behavior fitness value, and explores the solution space in the time domain for the time domain behavior fitness value to obtain a first fault prediction result.
[0055] Optionally, the inputting the second characteristic parameter into the constructed fault prediction model to perform fault prediction to obtain a second fault prediction result includes:
[0056] Inputting the second characteristic parameter into the constructed fault prediction model to perform Levy motion and Brownian motion analysis at multiple frequency scales based on the second characteristic parameter to obtain a Levy motion analysis frequency sequence and a Brownian motion analysis frequency sequence;
[0057] Based on the Levy motion analysis frequency sequence and the Brownian motion analysis frequency sequence, fault prediction is performed on the vulnerable component to obtain a second fault prediction result.
[0058] Optionally, the step of inputting the second characteristic parameter into a constructed fault prediction model to perform multi-frequency scale Levy motion and Brownian motion analysis based on the second characteristic parameter to obtain a Levy motion analysis frequency sequence and a Brownian motion analysis frequency sequence comprises:
[0059] Based on the constructed fault prediction model, long-tail distribution and self-similarity of the second characteristic parameter at multiple frequency scales are counted to perform Levy motion analysis at multiple frequency scales on the second characteristic parameter to obtain a Levy motion analysis frequency sequence;
[0060] Based on the constructed fault prediction model, the random motion of the second characteristic parameter on multiple frequency scales is tracked to perform Brownian motion analysis on the second characteristic parameter to obtain a Brownian motion analysis frequency sequence.
[0061] Optionally, the performing fault prediction on the vulnerable component based on the Levy motion analysis frequency sequence and the Brownian motion analysis frequency sequence to obtain a second fault prediction result includes:
[0062] Vectorizing the Levy motion analysis frequency sequence and the Brownian motion analysis frequency sequence respectively to obtain a Levy motion feature vector and a Brownian motion feature vector;
[0063] Performing vector fusion on the Levy motion feature vector and the Brownian motion feature vector to obtain a fault prediction feature vector;
[0064] The constructed fault prediction model performs walking iteration on the initial fault prediction parameters constructed for the vulnerable part based on the fault prediction feature vector, so as to calculate the frequency domain development behavior fitness of the vulnerable part to obtain a frequency domain behavior fitness value, and explores the solution space in the frequency domain for the frequency domain behavior fitness value to obtain a second fault prediction result.
[0065] Optionally, the fusing the first fault prediction result and the second fault prediction result to obtain a joint fault prediction result for the vulnerable component includes:
[0066] Based on the set weights, a weighted operation is performed on the probabilities corresponding to the first fault prediction result and the second fault prediction result to obtain the probability of the joint fault prediction result for the vulnerable component.
[0067] Optionally, performing fault prediction on the vulnerable component based on the joint fault prediction result includes: matching a historical fault statistics table of the vulnerable component with the probability of the joint fault prediction result to evaluate the confidence of the probability of the joint fault prediction result, and performing fault prediction on the vulnerable component based on the confidence.
[0068] In the present invention, the method for predicting the fault of vulnerable parts in the logistics equipment of the above claims is to extract the first feature and the second feature of the real-time status monitoring data of the vulnerable parts to obtain feature parameters from multiple angles. This multi-feature extraction method can more comprehensively reflect the operating status of the vulnerable parts and capture more factors that may affect their performance. Compared with the traditional method that relies only on a single feature parameter for prediction, multi-feature extraction significantly improves the comprehensiveness and accuracy of the prediction. In addition, after predicting the fault based on the first feature parameter and the second feature parameter respectively, the method fuses the two prediction results to obtain a joint fault prediction result. This fusion strategy makes full use of the complementarity and correlation between different feature parameters to further improve the accuracy and reliability of the prediction. By comprehensively considering multiple prediction results, the error and uncertainty that may be caused by a single prediction are reduced. Furthermore, the method can timely detect performance changes or potential faults of vulnerable parts by monitoring their operating status in real time. Compared with the traditional regular maintenance method, real-time monitoring greatly improves the timeliness of fault prediction, and can provide early warning and processing before or at the early stage of the fault, thereby avoiding the expansion of the fault and the increase of losses. Moreover, this method predicts faults based on real-time status monitoring data, achieving accurate grasp and prediction of the status of vulnerable parts. Compared with the traditional method based on experience judgment, the data-driven approach is more scientific and objective, reducing the interference and errors of human factors. In summary, by timely predicting and warning of the failure of vulnerable parts, this method can arrange maintenance and replacement work in advance and reduce downtime caused by equipment failure. This not only improves the operating efficiency of logistics equipment, but also ensures the continuity and stability of logistics operations. BRIEF DESCRIPTION OF THE DRAWINGS
[0069] Hereinafter, some specific embodiments of the present application will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings:
[0070] Figure 1 A schematic flow chart of a method for predicting failures of vulnerable parts in logistics equipment provided in an embodiment of the present application. DETAILED DESCRIPTION
[0071] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the embodiments of the present application should fall within the scope of protection of the embodiments of the present application.
[0072] Figure 1 A schematic diagram of a method for predicting a failure of a vulnerable component in a logistics device provided in an embodiment of the present application. Figure 1 As shown, it includes:
[0073] Performing real-time monitoring on the operating status of the vulnerable parts to obtain real-time status monitoring data of the vulnerable parts;
[0074] Performing a first feature extraction on the real-time status monitoring data to obtain a first feature parameter;
[0075] Performing a second feature extraction on the real-time status monitoring data to obtain a second feature parameter;
[0076] Performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result;
[0077] Performing fault prediction based on the second characteristic parameter to obtain a second fault prediction result;
[0078] fusing the first fault prediction result and the second fault prediction result to obtain a joint fault prediction result for the vulnerable component;
[0079] According to the combined fault prediction result, fault prediction is performed on the vulnerable component.
[0080] In the present invention, the method for predicting the fault of vulnerable parts in the logistics equipment of the above claims is to extract the first feature and the second feature of the real-time status monitoring data of the vulnerable parts to obtain feature parameters from multiple angles. This multi-feature extraction method can more comprehensively reflect the operating status of the vulnerable parts and capture more factors that may affect their performance. Compared with the traditional method that relies only on a single feature parameter for prediction, multi-feature extraction significantly improves the comprehensiveness and accuracy of the prediction. In addition, after predicting the fault based on the first feature parameter and the second feature parameter respectively, the method fuses the two prediction results to obtain a joint fault prediction result. This fusion strategy makes full use of the complementarity and correlation between different feature parameters to further improve the accuracy and reliability of the prediction. By comprehensively considering multiple prediction results, the error and uncertainty that may be caused by a single prediction are reduced. Furthermore, the method can timely detect performance changes or potential faults of vulnerable parts by monitoring their operating status in real time. Compared with the traditional regular maintenance method, real-time monitoring greatly improves the timeliness of fault prediction, and can provide early warning and processing before or at the early stage of the fault, thereby avoiding the expansion of the fault and the increase of losses. Moreover, this method predicts faults based on real-time status monitoring data, achieving accurate grasp and prediction of the status of vulnerable parts. Compared with the traditional method based on experience judgment, the data-driven approach is more scientific and objective, reducing the interference and errors of human factors. In summary, by timely predicting and warning of the failure of vulnerable parts, this method can arrange maintenance and replacement work in advance and reduce downtime caused by equipment failure. This not only improves the operating efficiency of logistics equipment, but also ensures the continuity and stability of logistics operations.
[0081] Optionally, the real-time monitoring of the operating status of the vulnerable component to obtain real-time status monitoring data of the vulnerable component includes:
[0082] Based on the sensor arranged on the vulnerable part, the operating state of the vulnerable part is monitored in real time to obtain equipment operating state data, and based on the monitored time series, a time series-based equipment operating state data set is formed;
[0083] Preprocessing the equipment operation status data set to obtain a fault prediction data set, wherein the preprocessing includes at least one of cleaning, denoising, and normalization;
[0084] A feature analysis is performed on the fault prediction data set to obtain real-time status monitoring data of the vulnerable parts.
[0085] To this end, the above process of real-time monitoring of the operating status of vulnerable parts, data preprocessing and feature analysis has the following technical benefits:
[0086] (1) Real-time monitoring and collection of equipment operating status data through sensors, combined with time series to form a data set, can ensure the timeliness and continuity of the data. The subsequent preprocessing steps (such as cleaning, denoising, and normalization) further improve the quality of the data, remove noise and outliers, and make subsequent feature extraction and fault prediction more accurate and reliable.
[0087] (2) Feature analysis of the preprocessed data set helps to gain a deeper understanding of the operating rules and failure modes of vulnerable parts. Through feature analysis, characteristic parameters that have an important impact on fault prediction can be extracted from complex operating data, providing richer information input for subsequent prediction models.
[0088] (3) Due to the use of time series-based data sets and multiple preprocessing techniques, the trained fault prediction model can better adapt to the state changes of vulnerable parts under different working conditions and environments, thereby improving the generalization ability of the model and the practical application effect.
[0089] In addition, the following provides an exemplary code framework to illustrate how to perform real-time monitoring, data preprocessing, and feature analysis on the operating status of vulnerable parts:
[0090]
[0091]
[0092] Optionally, the performing feature analysis on the fault prediction data set to obtain real-time status monitoring data of the vulnerable component therefrom includes:
[0093] A correlation / redundancy analysis is performed on the fault prediction data set to obtain real-time status monitoring data of the vulnerable component therefrom.
[0094] To this end, the above correlation / redundancy analysis of the fault prediction data set has the following technical benefits:
[0095] (1) Through correlation analysis, features that have a significant impact on fault prediction results can be identified, while features that have low correlation with the prediction target or are redundant can be eliminated. This can not only reduce the complexity of the model and improve computational efficiency, but also improve the prediction accuracy and generalization ability of the model.
[0096] (2) Redundant features (i.e., those that are highly correlated with other features or contain the same information) may cause overfitting during model training. By analyzing the redundancy and removing these features, the risk of overfitting can be effectively reduced, making the model more robust.
[0097] (3) Selecting highly relevant and non-redundant features for modeling can make the model's decision-making process more transparent and explainable. This is very important for understanding how the model makes predictions, diagnosing potential problems, and performing subsequent model optimization.
[0098] In addition, an exemplary code framework is provided below to illustrate the process of performing correlation / redundancy analysis on a fault prediction data set and obtaining optimized real-time condition monitoring data (features) therefrom:
[0099]
[0100]
[0101] Optionally, the correlation analysis is used to analyze the correlation between the fault prediction data set and the state of the wearing parts to obtain a correlation measurement value, so as to select the equipment operation status data with a correlation measurement value greater than a set correlation threshold as the alternative status monitoring data to form an alternative status monitoring data set; the redundancy analysis is used to analyze the semantic similarity of the alternative status monitoring data in the alternative status monitoring data set to obtain a similarity measurement value, so as to deduplicate the alternative status monitoring data with a similarity measurement value greater than a set similarity threshold to obtain the real-time status monitoring data.
[0102] Therefore, the specific application process based on the above correlation analysis and redundancy analysis has the following technical benefits:
[0103] (1) Correlation analysis helps identify equipment operating status data that is highly correlated with the status of vulnerable parts. These highly correlated data usually contain information that has a significant impact on the prediction results, so using them as alternative condition monitoring data can significantly improve the accuracy of fault prediction.
[0104] (2) Redundancy analysis: By performing semantic similarity analysis on the data in the candidate state monitoring dataset, we identify and remove data whose similarity metric exceeds the set threshold. This can reduce redundant information in the dataset, avoid overfitting problems caused by redundant features during model training, and reduce computational complexity and storage requirements.
[0105] (3) By combining correlation analysis with redundancy analysis, a real-time status monitoring data set can be constructed that contains key information while avoiding redundancy. When used for model training, such a data set can significantly improve the training efficiency and prediction speed of the model because the model no longer needs to process a large amount of useless or duplicate data.
[0106] (4) Using selected and de-redundant features to build a model can make the model's decision-making process more transparent and explainable. In addition, after removing redundant features, the model's prediction results are usually more robust and less susceptible to interference from noisy data.
[0107] (5) The optimization of the real-time condition monitoring data set enables the model to more accurately capture the early signs of failure of vulnerable parts. This is crucial for early diagnosis and prevention of failures, helping to reduce economic losses and safety hazards caused by downtime.
[0108] Optionally, performing a first feature extraction on the real-time status monitoring data to obtain a first feature parameter includes:
[0109] In the time domain, time domain feature extraction is performed on the real-time status monitoring data to obtain time domain feature parameters.
[0110] Optionally, in the time domain, extracting time domain features from the real-time status monitoring data to obtain time domain feature parameters includes:
[0111] Based on the monitored time series, the real-time state monitoring data is sampled in the order of the real-time state monitoring data to obtain real-time state time domain sampling data to form a real-time state time domain sampling data sequence;
[0112] Based on the set feature extraction index, feature extraction is performed on the real-time state time domain sampling data sequence to obtain time domain feature parameters.
[0113] To this end, the above-mentioned feature extraction of real-time status monitoring data in the time domain to obtain time domain feature parameters to capture dynamic changes has the following technical benefits:
[0114] (1) Feature extraction of real-time status monitoring data in the time domain can directly capture the dynamic characteristics of equipment status over time. This is particularly important for predicting failure modes that are closely related to time, such as wear and fatigue, which gradually appear as the operating time increases.
[0115] (2) Time domain feature parameters usually contain rich information and can more comprehensively reflect the operating status of the equipment. By extracting these features, the model can learn more details about the changes in equipment status, thereby improving the accuracy of fault prediction.
[0116] (3) Since time domain features are extracted based on time series data, they can reflect the state changes of the equipment at different time points. This feature enables the model to learn more general feature representations during training, thereby enhancing the model's generalization ability under different working conditions.
[0117] (4) Feature extraction in the time domain is relatively intuitive and simple, and does not require complex transformations or conversions. This helps to simplify the data processing process, reduce data processing costs, and improve data processing efficiency.
[0118] In addition, the following exemplary code is provided to illustrate how to perform feature extraction on real-time condition monitoring data in the time domain to obtain time domain feature parameters:
[0119]
[0120]
[0121] Optionally, the extracting features of the real-time state time domain sampling data sequence based on the set feature extraction index to obtain time domain feature parameters includes:
[0122] Based on the set dimensional index, the semantic feature extraction of the actual physical meaning of the real-time state time domain sampling data sequence is performed to obtain a first time domain feature parameter;
[0123] Based on the set dimensionless index, the real-time state time domain sampling data sequence is distributedly sampled to obtain a second time domain characteristic parameter.
[0124] Optionally, the first time domain characteristic parameter includes: at least one of: mean value, variance, peak-to-peak value, root mean square value, root amplitude, average amplitude, and peak value; the second time domain characteristic parameter includes: at least one of: slope index, kurtosis index, peak factor, margin index, waveform factor, and pulse index.
[0125] To this end, the feature extraction of the real-time state time domain sampling data sequence based on dimensional and dimensionless indicators has the following technical advantages:
[0126] (1) Combining dimensional and dimensionless feature extraction indicators can more comprehensively reflect the operating status of the equipment. Dimensional features (such as mean and variance) provide the actual physical meaning of the equipment status, while dimensionless features (such as slope and kurtosis) describe the morphology and statistical characteristics of data distribution. The two complement each other and jointly improve the accuracy and reliability of fault prediction.
[0127] (2) Dimensionless features do not depend on the absolute level or amplitude of the signal, so they are more robust to noise and interference in the signal. This means that even if there is a certain amount of noise or fluctuation in the actual monitoring data, the model can still stably extract useful feature information, thereby improving the stability and reliability of the model.
[0128] (3) By extracting multiple types of time-domain feature parameters, the model can learn more details about the changes in equipment status. These rich feature information helps the model more accurately identify abnormalities or failure modes in equipment status, thereby improving the accuracy and timeliness of fault prediction.
[0129] (4) Different devices and application scenarios may require attention to different types of features. By providing a variety of dimensional and dimensionless feature extraction indicators, the solution can flexibly adapt to different application requirements and provide strong support for various equipment status monitoring and fault prediction tasks.
[0130] In addition, the following is an example code to illustrate how to extract features from a real-time state time domain sampling data sequence based on dimensional and dimensionless indicators:
[0131]
[0132]
[0133] Optionally, performing a second feature extraction on the real-time status monitoring data to obtain a second feature parameter includes:
[0134] Based on the time-frequency transformation model, the real-time status monitoring data is transformed from the time domain to the frequency domain to obtain frequency domain status monitoring data, and frequency domain features are extracted from the frequency domain status monitoring data to obtain frequency domain feature parameters.
[0135] Optionally, the time-frequency transformation model is a time-frequency transformation model, a Fourier transformation model and / or a Fourier series decomposition model.
[0136] Optionally, the extracting frequency domain features from the frequency domain state monitoring data to obtain frequency domain feature parameters includes:
[0137] Based on the set periodicity and boundedness constraints, the frequency data of the frequency domain state monitoring data is determined, and frequency domain features of the frequency domain state monitoring data are extracted based on the frequency data, and a correlation relationship between the extracted frequency domain features and the frequency data is established to obtain the frequency domain feature parameters.
[0138] To this end, the above-mentioned time-frequency transformation and frequency domain feature extraction of real-time status monitoring data have the following technical benefits:
[0139] (1) Through time-frequency transformation models (such as Fourier transform or Fourier series decomposition), real-time condition monitoring data can be converted from the time domain to the frequency domain, thereby revealing possible periodic components in the data. This is particularly useful for diagnosing vibration problems in rotating machinery (such as motors and bearings) because many failure modes exhibit specific frequency characteristics.
[0140] (2) Frequency domain feature extraction can capture subtle changes that may not be easily perceived in the time domain. For example, some faults may appear as non-periodic random fluctuations in the time domain, but can clearly present specific frequency peaks in the frequency domain. This helps to more accurately identify and locate the source of the fault.
[0141] (3) Frequency domain features usually have a certain inhibitory effect on noise and interference in the signal. This is because noise and interference may be distributed over the entire spectrum in the frequency domain, while fault features are often concentrated in a specific frequency range. By extracting these specific frequency domain features, the impact of noise on the diagnosis results can be reduced.
[0142] (4) In complex systems, data from multiple sensors may be monitored simultaneously. By converting all this data into the frequency domain and extracting their respective frequency domain features, it is easier to fuse information from different sensors to obtain a more comprehensive description of the system status.
[0143] In addition, the following is an exemplary code to illustrate how to perform time-frequency transformation and frequency domain feature extraction on real-time condition monitoring data:
[0144]
[0145]
[0146] Optionally, establishing an association relationship between the extracted frequency domain features and the frequency data to obtain the frequency domain feature parameters includes:
[0147] Based on the mapping model, the frequency domain characteristics and frequency data are fitted with a correlation relationship to obtain the frequency domain characteristic parameters.
[0148] Optionally, the frequency domain characteristic parameter includes: at least one of: centroid frequency, mean square frequency, root mean square frequency, mean square deviation frequency, and standard deviation frequency.
[0149] Optionally, the mapping model includes at least one of a power spectrum model, a phase spectrum model, an amplitude spectrum model, and an envelope spectrum model.
[0150] To this end, the above-mentioned extraction of frequency domain feature parameters based on the mapping model (taking the power spectrum model as an example) and the establishment of the correlation relationship between the frequency domain features and the frequency data have the following technical benefits:
[0151] (1) By establishing the correlation between frequency domain features and frequency data, the characteristics of the signal in the frequency domain, such as energy distribution, main frequency components, etc., can be quantitatively described. This is of great significance for applications such as fault identification, performance evaluation, and pattern classification.
[0152] (2) Mapping models (such as power spectrum models, phase spectrum models, etc.) can accurately capture subtle changes in the signal in the frequency domain, thereby improving the accuracy and reliability of the analysis. This is especially important for application scenarios that require high-precision analysis.
[0153] (3) Frequency domain characteristic parameters (such as center of gravity frequency, mean square frequency, etc.) provide the statistical characteristics of the signal in the frequency domain, which helps to more comprehensively understand the operating status of the equipment. These parameters can serve as an important basis for fault diagnosis, performance prediction and condition monitoring.
[0154] (4) Different frequency domain feature parameters can describe the frequency domain characteristics of the signal from different perspectives. By fusing multiple frequency domain feature parameters, a more comprehensive fault identification model can be constructed to improve the accuracy and robustness of diagnosis.
[0155] In addition, an exemplary code is provided below to illustrate how to extract frequency domain feature parameters based on a mapping model (taking a power spectrum model as an example) and establish an association relationship between frequency domain features and frequency data:
[0156]
[0157]
[0158] Optionally, the performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result includes: predicting an operating state of the vulnerable component based on the first characteristic parameter to obtain a first predicted operating state, and diagnosing a fault of the vulnerable component based on the first predicted operating state to obtain a first fault prediction result;
[0159] The performing fault prediction based on the second characteristic parameter to obtain a second fault prediction result includes: predicting the operating state of the wearing part based on the second characteristic parameter to obtain a second predicted operating state, and diagnosing the fault of the wearing part based on the second predicted operating state to obtain a second fault prediction result.
[0160] To this end, the technical solution here has the following technical benefits:
[0161] (1) By combining the first characteristic parameter and the second characteristic parameter for fault prediction, the operating status of vulnerable parts can be comprehensively evaluated from different angles, thereby improving the accuracy of prediction.
[0162] (2) By predicting the operating status of vulnerable parts, potential problems can be detected before failure occurs, achieving early warning, thereby avoiding or reducing downtime and maintenance costs caused by failure.
[0163] (3) Fault diagnosis based on the predicted operating status can locate the fault source more accurately, provide strong support for subsequent maintenance work, and improve maintenance efficiency and accuracy.
[0164] (4) Based on the fault prediction results, a more scientific and reasonable maintenance plan can be formulated, such as preventive maintenance, predictive maintenance, etc., to reduce maintenance costs and improve equipment reliability and service life.
[0165] In addition, the following provides an example code to illustrate how to use different feature parameters:
[0166]
[0167]
[0168] Optionally, predicting the operating state of the consumable part based on the first characteristic parameter to obtain a first predicted operating state includes: loading the first characteristic parameter onto a digital twin model of the consumable part to perform a simulation operation to predict the operating state of the consumable part to obtain the first predicted operating state.
[0169] The predicting the operating state of the consumable component based on the second characteristic parameter to obtain a second predicted operating state includes: loading the second characteristic parameter onto a digital twin model of the consumable component to perform a simulation operation to predict the operating state of the consumable component to obtain a second predicted operating state.
[0170] To this end, loading characteristic parameters into the digital twin model for simulation and prediction has the following technical benefits:
[0171] (1) By using the digital twin model for simulation operation, the state of vulnerable parts in the actual operating environment can be highly restored, thereby providing more accurate prediction results.
[0172] (2) The digital twin model can receive and process data from sensors (i.e., the first and second characteristic parameters) in real time, thus realizing real-time monitoring and prediction of the operating status of vulnerable parts.
[0173] (3) The digital twin model allows repeated simulation of different operating conditions and fault scenarios in a virtual environment, providing rich data support for fault prediction and fault diagnosis.
[0174] (4) Various types of characteristic parameters (such as frequency domain, time domain, statistical characteristics, etc.) can be easily integrated into the digital twin model for comprehensive analysis and prediction, thereby improving the comprehensiveness and accuracy of the prediction.
[0175] In addition, an example code is provided below to illustrate how to load feature parameters into the digital twin model for simulation and prediction.
[0176]
[0177]
[0178] Optionally, performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result includes:
[0179] The first characteristic parameter is input into the constructed fault prediction model to perform fault prediction and obtain a first fault prediction result.
[0180] Optionally, inputting the first characteristic parameter into a constructed fault prediction model to perform fault prediction to obtain a first fault prediction result includes:
[0181] Inputting the first characteristic parameter into the constructed fault prediction model to perform multi-time scale Levy motion and Brownian motion analysis based on the first characteristic parameter to obtain a Levy motion analysis time series and a Brownian motion analysis time series;
[0182] Based on the Levy motion analysis time series and the Brownian motion analysis time series, fault prediction is performed on the vulnerable component to obtain a first fault prediction result.
[0183] To this end, the first characteristic parameter is input into the constructed fault prediction model, and multi-time scale Levy motion and Brownian motion analysis is performed to perform fault prediction, which has the following technical benefits:
[0184] (1) By introducing multi-time scale Lévy motion and Brownian motion analysis, the operating state changes of vulnerable parts at different time scales can be captured more carefully, thereby improving the accuracy and sensitivity of fault prediction.
[0185] (2) Lévy motion and Brownian motion represent random walk and diffusion phenomena, respectively, which have a wide range of applications in physical systems. Introducing these two motion modes into the fault prediction model can achieve an organic combination of physical phenomena and statistical methods, and improve the model's interpretability and predictive capabilities.
[0186] (3) Multi-time scale analysis helps to detect early signs of failure, thus providing enterprises with longer response time so that they can take preventive maintenance measures to avoid failures or reduce the losses caused by failures.
[0187] (4) The fault prediction model can be adjusted and optimized according to actual needs to adapt to the requirements of different vulnerable parts and operating environments. At the same time, the model can also be expanded to include more characteristic parameters and analysis methods to further improve the prediction performance.
[0188] In addition, the following is a sample code to illustrate how to input the first characteristic parameter into the constructed fault prediction model and perform multi-time scale Levy motion and Brownian motion analysis for fault prediction:
[0189]
[0190] Optionally, the step of inputting the first characteristic parameter into a constructed fault prediction model to perform multi-time-scale Levy motion and Brownian motion analysis based on the first characteristic parameter to obtain a Levy motion analysis time series and a Brownian motion analysis time series comprises:
[0191] Based on the constructed fault prediction model, long-tail distribution and self-similarity of the first characteristic parameter at multiple time scales are statistically analyzed to perform Levy motion analysis at multiple time scales on the first characteristic parameter to obtain a Levy motion analysis time series;
[0192] Based on the constructed fault prediction model, the random motion of the first characteristic parameter on multiple time scales is tracked to perform Brownian motion analysis on the first characteristic parameter to obtain a Brownian motion analysis time series.
[0193] To this end, the above-mentioned multi-time scale Lévy motion and Brownian motion analysis based on the constructed fault prediction model and the generation of corresponding analysis time series have the following technical benefits:
[0194] (1) Through the analysis of Lévy motion and Brownian motion at multiple time scales, we can gain a deeper understanding of the dynamic behavior of vulnerable components during operation, including their randomness and self-similarity, which helps to predict failures more accurately.
[0195] (2) Combining the analysis of long-tail distribution and self-similarity can more accurately capture the changing trend of system status, thereby improving the accuracy and reliability of fault prediction.
[0196] (3) Through multi-time scale analysis, small changes in system status can be discovered earlier, which may be precursors to failures, thus providing enterprises with longer response time for preventive maintenance.
[0197] (4) Since the characteristics of multiple time scales are taken into account, the model is less sensitive to slight changes or noise in the input data, thus improving the robustness and stability of the model.
[0198] In addition, the following is a sample code to illustrate how to perform multi-time scale Levy motion and Brownian motion analysis based on the constructed fault prediction model and generate the corresponding analysis time series:
[0199]
[0200]
[0201] Optionally, the performing fault prediction on the vulnerable component based on the Levy motion analysis time series and the Brownian motion analysis time series to obtain a first fault prediction result includes:
[0202] Vectorizing the Levy motion analysis time series and the Brownian motion analysis time series respectively to obtain a Levy motion feature vector and a Brownian motion feature vector;
[0203] Performing vector fusion on the Levy motion feature vector and the Brownian motion feature vector to obtain a fault prediction feature vector;
[0204] The constructed fault prediction model performs walking iteration on the initial fault prediction parameters constructed for the vulnerable component based on the fault prediction feature vector, so as to calculate the time domain development behavior fitness of the vulnerable component to obtain a time domain behavior fitness value, and explores the solution space in the time domain for the time domain behavior fitness value to obtain a first fault prediction result.
[0205] To this end, feature fusion, time domain development behavior fitness calculation and first fault prediction result generation based on Levy motion analysis time series and Brownian motion analysis time series have the following technical benefits:
[0206] (1) By converting the Levy motion analysis time series and the Brownian motion analysis time series into feature vectors and fusing them, the advantages of the two analysis methods can be combined to capture more dimensions of system dynamic information, thereby improving the accuracy and comprehensiveness of fault prediction.
[0207] (2) Based on the fault prediction feature vector and through the calculation of the time domain evolution behavior fitness, the model can simulate the behavioral changes of vulnerable parts during actual operation and more accurately predict their status at different time points, including potential fault points.
[0208] (3) In addition to calculating the time domain behavior fitness, the model also explores the solution space in the time domain. This means that it can not only predict the current or near-term failure risks, but also predict the system status changes over a longer period of time in the future to a certain extent, providing enterprises with a longer warning period and decision-making support.
[0209] (4) The iterative optimization of the initial fault prediction parameters using the walking iteration algorithm can improve the model's adaptability and prediction accuracy, enabling the model to better adapt to the complex and changeable system operating environment.
[0210] In addition, the following is a sample code to illustrate how to perform feature fusion, calculate the fitness of time-domain development behavior, and generate the first fault prediction result based on the Levy motion analysis time series and the Brownian motion analysis time series:
[0211]
[0212]
[0213] Optionally, the inputting the second characteristic parameter into the constructed fault prediction model to perform fault prediction to obtain a second fault prediction result includes:
[0214] Inputting the second characteristic parameter into the constructed fault prediction model to perform Levy motion and Brownian motion analysis at multiple frequency scales based on the second characteristic parameter to obtain a Levy motion analysis frequency sequence and a Brownian motion analysis frequency sequence;
[0215] Based on the Levy motion analysis frequency sequence and the Brownian motion analysis frequency sequence, fault prediction is performed on the vulnerable component to obtain a second fault prediction result.
[0216] To this end, the above-mentioned Lévy motion and Brownian motion analysis based on multiple frequency scales has the following technical benefits:
[0217] (1) By introducing multi-frequency scale Lévy motion and Brownian motion analysis, the technology can capture the dynamic behavior characteristics of vulnerable parts at different frequency scales. This multi-scale analysis capability helps to more fully understand the failure mechanism of the components, thereby improving the accuracy and reliability of fault prediction.
[0218] (2) Lévy motion and Brownian motion represent the physical processes of random walk and molecular thermal motion, respectively. They are often used in the engineering field to simulate and analyze the dynamic behavior of complex systems. Applying these physical concepts to fault prediction can make the prediction results have clearer physical meanings and facilitate understanding and interpretation.
[0219] (3) Since Lévy motion and Brownian motion analysis do not rely on specific system models or assumptions, the technology has strong adaptability. It can be applied to different types of vulnerable parts and failure modes, providing strong support for failure prediction of various engineering systems.
[0220] (4) By combining the results of Levy motion and Brownian motion analysis, the performance changes of components at different frequency scales can be comprehensively considered, so as to more accurately predict their failure trends. This comprehensive analysis method helps to improve prediction accuracy and reduce false positives and false negatives.
[0221] In addition, the following example code is provided to illustrate how to apply this analysis method to the fault prediction model:
[0222]
[0223] Optionally, the step of inputting the second characteristic parameter into a constructed fault prediction model to perform multi-frequency scale Levy motion and Brownian motion analysis based on the second characteristic parameter to obtain a Levy motion analysis frequency sequence and a Brownian motion analysis frequency sequence comprises:
[0224] Based on the constructed fault prediction model, long-tail distribution and self-similarity of the second characteristic parameter at multiple frequency scales are counted to perform Levy motion analysis at multiple frequency scales on the second characteristic parameter to obtain a Levy motion analysis frequency sequence;
[0225] Based on the constructed fault prediction model, the random motion of the second characteristic parameter on multiple frequency scales is tracked to perform Brownian motion analysis on the second characteristic parameter to obtain a Brownian motion analysis frequency sequence.
[0226] To this end, processing the second characteristic parameter based on multi-frequency scale Levy motion and Brownian motion analysis and generating the corresponding analysis frequency sequence has the following technical benefits:
[0227] (1) Through the analysis of Lévy motion and Brownian motion at multiple frequency scales, the dynamic behavior characteristics of vulnerable components at different frequencies can be more comprehensively captured. This is particularly important for understanding the failure mechanism of complex systems, because the system response at different frequencies may reveal different failure modes and early warning signs.
[0228] (2) Combining the analysis of multiple frequency scales can more effectively extract the key information in the characteristic parameters and reduce the interference of noise and redundant data, thereby improving the accuracy and reliability of fault prediction.
[0229] (3) Analyzing the dynamic behavior of the system in the frequency domain can reveal features that are difficult to directly observe in the time domain, such as resonant frequency, spectral characteristics, etc. These features are important for identifying specific types of faults and predicting their development trends.
[0230] (4) By considering the characteristics of multiple frequency scales, the constructed fault prediction model can better adapt to system changes under different working conditions and operating conditions, and improve the generalization ability and robustness of the model.
[0231] In addition, the following is a sample code to illustrate how to process the second characteristic parameter based on the Lévy motion and Brownian motion analysis at multiple frequency scales and generate the corresponding analysis frequency sequence:
[0232]
[0233] Optionally, the performing fault prediction on the vulnerable component based on the Levy motion analysis frequency sequence and the Brownian motion analysis frequency sequence to obtain a second fault prediction result includes:
[0234] Vectorizing the Levy motion analysis frequency sequence and the Brownian motion analysis frequency sequence respectively to obtain a Levy motion feature vector and a Brownian motion feature vector;
[0235] Performing vector fusion on the Levy motion feature vector and the Brownian motion feature vector to obtain a fault prediction feature vector;
[0236] The constructed fault prediction model performs walking iteration on the initial fault prediction parameters constructed for the vulnerable part based on the fault prediction feature vector, so as to calculate the frequency domain development behavior fitness of the vulnerable part to obtain a frequency domain behavior fitness value, and explores the solution space in the frequency domain for the frequency domain behavior fitness value to obtain a second fault prediction result.
[0237] The above-mentioned walking iteration algorithm and frequency domain solution space exploration process have the following technical benefits:
[0238] (1) Converting the Levy motion analysis frequency sequence and the Brownian motion analysis frequency sequence into feature vectors and fusing them can integrate the feature information of the two different motion modes, thereby more comprehensively describing the dynamic behavior characteristics of vulnerable parts. This fusion method helps capture complex failure modes and improve the accuracy of failure prediction.
[0239] (2) Calculating the fitness of the development behavior of vulnerable components in the frequency domain can more accurately assess the health status and potential failure risk of components at different frequencies. This method can capture subtle changes in the frequency domain and provide strong support for early fault warning.
[0240] (3) By exploring the solution space of the frequency domain behavior fitness value, we can further explore the potential fault modes and fault development trends. This exploration process helps to discover new fault features and improve the flexibility and robustness of fault prediction.
[0241] (4) The constructed fault prediction model uses a walking iteration algorithm to update and optimize the initial fault prediction parameters, which can gradually approach the actual fault state. This iterative method can make full use of historical data and real-time monitoring information to improve the real-time and accuracy of fault prediction.
[0242] In addition, an exemplary code is provided below to illustrate the above-mentioned walking iteration algorithm and the frequency domain solution space exploration process:
[0243]
[0244]
[0245] Optionally, the fusing the first fault prediction result and the second fault prediction result to obtain a joint fault prediction result for the vulnerable component includes:
[0246] Based on the set weights, a weighted operation is performed on the probabilities corresponding to the first fault prediction result and the second fault prediction result to obtain the probability of the joint fault prediction result for the vulnerable component.
[0247] To this end, the solution here has the following technical benefits:
[0248] (1) By fusing the first fault prediction result and the second fault prediction result, the advantages of different prediction methods and models can be comprehensively considered to obtain a more accurate joint fault prediction result. This method can reduce the deviation and uncertainty that may be caused by a single prediction method.
[0249] (2) Since the joint fault prediction result is based on the fusion of multiple prediction results, even if a prediction method has errors or fails, the overall prediction result can still maintain a certain degree of accuracy. This fusion method enhances the robustness and reliability of the prediction system.
[0250] (3) By setting different weights, the contribution of different prediction results in the joint prediction can be adjusted according to the actual situation. This flexibility enables the prediction system to adapt to different application scenarios and demand changes.
[0251] (4) Weighted operation is a relatively simple and easy-to-implement fusion method that does not require complex algorithms and model design. This makes the joint fault prediction results more convenient and efficient in practical applications.
[0252] In addition, an exemplary code is provided below to illustrate the above implementation process:
[0253]
[0254]
[0255] #The output result will show the joint fault prediction probability after the two prediction results are fused according to the set weights
[0256] Optionally, performing fault prediction on the vulnerable component based on the joint fault prediction result includes: matching a historical fault statistics table of the vulnerable component with the probability of the joint fault prediction result to evaluate the confidence of the probability of the joint fault prediction result, and performing fault prediction on the vulnerable component based on the confidence.
[0257] To this end, the confidence of the prediction results is evaluated based on the combined fault prediction results and the historical failure statistics table of vulnerable parts, and fault prediction is performed, which has the following technical benefits:
[0258] (1) By matching the probability of the joint fault prediction result with the historical failure statistics of the vulnerable parts, the confidence of the prediction result can be evaluated. This method uses historical data to verify the reliability of the prediction result, thereby improving the confidence and accuracy of the fault prediction.
[0259] (2) Since the failure modes and frequencies of different vulnerable parts may be different, more personalized failure prediction can be achieved by combining their historical failure statistics. This method can more accurately reflect the actual failure conditions of vulnerable parts and improve the pertinence and effectiveness of prediction.
[0260] (3) By continuously collecting and analyzing the failure data of vulnerable parts, the historical failure statistics table can be updated and improved, thereby optimizing the failure prediction model. This continuous improvement mechanism helps maintain the accuracy and timeliness of the prediction system.
[0261] (4) After evaluating the confidence level of the joint fault prediction results, it can provide decision makers with a more reliable and scientific basis to formulate targeted maintenance plans and response strategies, thereby avoiding or reducing losses caused by failures.
[0262] In addition, the following is an exemplary code for illustrating how to evaluate the confidence of the prediction results and perform failure prediction based on the joint failure prediction results and the historical failure statistics table of the vulnerable parts:
[0263]
[0264]
[0265] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with the technical features with similar functions disclosed in this application (but not limited to) by each other to form a technical solution.
Claims
1. A method for predicting failure of vulnerable parts in logistics equipment, characterized in that: include: Performing real-time monitoring on the operating status of the vulnerable parts to obtain real-time status monitoring data of the vulnerable parts; Performing a first feature extraction on the real-time status monitoring data to obtain a first feature parameter; Performing a second feature extraction on the real-time status monitoring data to obtain a second feature parameter; Performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result; Performing fault prediction based on the second characteristic parameter to obtain a second fault prediction result; fusing the first fault prediction result and the second fault prediction result to obtain a joint fault prediction result for the vulnerable component; According to the combined fault prediction result, fault prediction is performed on the vulnerable component. The real-time monitoring of the operating status of the vulnerable parts to obtain real-time status monitoring data of the vulnerable parts includes: Based on the sensor arranged on the vulnerable part, the operating state of the vulnerable part is monitored in real time to obtain equipment operating state data, and based on the monitored time series, a time series-based equipment operating state data set is formed; Preprocessing the equipment operation status data set to obtain a fault prediction data set, wherein the preprocessing includes at least one of cleaning, denoising, and normalization; Performing feature analysis on the fault prediction data set to obtain real-time status monitoring data of the vulnerable parts; The performing of feature analysis on the fault prediction data set to obtain real-time status monitoring data of the vulnerable parts therefrom includes: A correlation / redundancy analysis is performed on the fault prediction data set to obtain real-time status monitoring data of the vulnerable component therefrom.
2. A method for predicting failure of vulnerable parts in logistics equipment according to claim 1, characterized in that: The correlation analysis is used to analyze the correlation between the fault prediction data set and the state of the wearing parts to obtain a correlation measurement value, so as to select the equipment operation status data whose correlation measurement value is greater than the set correlation threshold as the alternative status monitoring data to form the alternative status monitoring data set; the redundancy analysis is used to analyze the semantic similarity of the alternative status monitoring data in the alternative status monitoring data set to obtain a similarity measurement value, so as to perform redundancy processing on the alternative status monitoring data whose similarity measurement value is greater than the set similarity threshold to obtain the real-time status monitoring data.
3. A method for predicting failure of vulnerable parts in logistics equipment according to claim 1, characterized in that: The performing a first feature extraction on the real-time status monitoring data to obtain a first feature parameter includes: In the time domain, time domain feature extraction is performed on the real-time status monitoring data to obtain time domain feature parameters.
4. A method for predicting failure of vulnerable parts in logistics equipment according to claim 3, characterized in that: In the time domain, extracting time domain features from the real-time status monitoring data to obtain time domain feature parameters includes: Based on the monitored time series, the real-time state monitoring data is sampled in the order of the real-time state monitoring data to obtain real-time state time domain sampling data to form a real-time state time domain sampling data sequence; Based on the set feature extraction index, feature extraction is performed on the real-time state time domain sampling data sequence to obtain time domain feature parameters; The step of extracting features from the real-time state time domain sampling data sequence based on the set feature extraction index to obtain time domain feature parameters includes: Based on the set dimensional index, the semantic feature extraction of the actual physical meaning of the real-time state time domain sampling data sequence is performed to obtain a first time domain feature parameter; Based on the set dimensionless index, the real-time state time domain sampling data sequence is distributedly sampled to obtain a second time domain characteristic parameter; Among them, the first time domain characteristic parameters include: at least one of the mean value, variance, peak-to-peak value, root mean square value, root amplitude, average amplitude, and peak value; the second time domain characteristic parameters include: at least one of the slope index, kurtosis index, peak factor, margin index, waveform factor, and pulse index.
5. The method for predicting failure of vulnerable parts in logistics equipment according to claim 1, characterized in that: The performing a second feature extraction on the real-time status monitoring data to obtain a second feature parameter includes: Based on the time-frequency transformation model, the real-time status monitoring data is transformed from the time domain to the frequency domain to obtain frequency domain status monitoring data, and frequency domain feature extraction is performed on the frequency domain status monitoring data to obtain frequency domain feature parameters; Wherein, the time-frequency transformation model is a time-frequency transformation model, a Fourier transformation model and / or a Fourier series decomposition model.
6. A method for predicting failure of vulnerable parts in logistics equipment according to claim 5, characterized in that: The extracting frequency domain features of the frequency domain state monitoring data to obtain frequency domain feature parameters includes: Based on the set periodicity and boundedness constraints, determine the frequency data of the frequency domain state monitoring data, extract the frequency domain features of the frequency domain state monitoring data based on the frequency data, and establish a correlation relationship between the extracted frequency domain features and the frequency data to obtain the frequency domain feature parameters; The step of establishing a correlation relationship between the extracted frequency domain features and the frequency data to obtain the frequency domain feature parameters includes: Based on the mapping model, the frequency domain characteristics and frequency data are fitted with a correlation relationship to obtain the frequency domain characteristic parameters.
7. A method for predicting failure of vulnerable parts in logistics equipment according to claim 6, characterized in that: The mapping model includes at least one of a power spectrum model, a phase spectrum model, an amplitude spectrum model, and an envelope spectrum model.
8. A method for predicting failure of vulnerable parts in logistics equipment according to claim 1, characterized in that: The performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result includes: predicting the operating state of the vulnerable component based on the first characteristic parameter to obtain a first predicted operating state, and diagnosing the fault of the vulnerable component based on the first predicted operating state to obtain a first fault prediction result; The performing fault prediction based on the second characteristic parameter to obtain a second fault prediction result includes: predicting the operating state of the wearing part based on the second characteristic parameter to obtain a second predicted operating state, and diagnosing the fault of the wearing part based on the second predicted operating state to obtain a second fault prediction result.
9. A method for predicting failure of vulnerable parts in logistics equipment according to claim 1, characterized in that: The performing fault prediction based on the first characteristic parameter to obtain a first fault prediction result includes: The first characteristic parameter is input into the constructed fault prediction model to perform fault prediction and obtain a first fault prediction result.
10. The method for predicting failure of vulnerable parts in logistics equipment according to claim 1, characterized in that: The step of inputting the first characteristic parameter into the constructed fault prediction model to perform fault prediction and obtain a first fault prediction result includes: Inputting the first characteristic parameter into the constructed fault prediction model to perform multi-time scale Levy motion and Brownian motion analysis based on the first characteristic parameter to obtain a Levy motion analysis time series and a Brownian motion analysis time series; Based on the Levy motion analysis time series and the Brownian motion analysis time series, fault prediction is performed on the vulnerable component to obtain a first fault prediction result; The step of inputting the first characteristic parameter into the constructed fault prediction model to perform multi-time scale Levy motion and Brownian motion analysis based on the first characteristic parameter to obtain a Levy motion analysis time series and a Brownian motion analysis time series includes: Based on the constructed fault prediction model, long-tail distribution and self-similarity of the first characteristic parameter at multiple time scales are statistically analyzed to perform Levy motion analysis at multiple time scales on the first characteristic parameter to obtain a Levy motion analysis time series; Based on the constructed fault prediction model, the random motion of the first characteristic parameter on multiple time scales is tracked to perform Brownian motion analysis on the first characteristic parameter to obtain a Brownian motion analysis time series; Wherein, the time series based on the Levy motion analysis and the Brownian motion analysis, Performing fault prediction on the vulnerable component to obtain a first fault prediction result includes: Vectorizing the Levy motion analysis time series and the Brownian motion analysis time series respectively to obtain a Levy motion feature vector and a Brownian motion feature vector; Performing vector fusion on the Levy motion feature vector and the Brownian motion feature vector to obtain a fault prediction feature vector; The constructed fault prediction model performs walking iteration on the initial fault prediction parameters constructed for the vulnerable component based on the fault prediction feature vector, so as to calculate the time domain development behavior fitness of the vulnerable component to obtain a time domain behavior fitness value, and explores the solution space in the time domain for the time domain behavior fitness value to obtain a first fault prediction result.
Citation Information
Cited By
Intelligent fault early warning method for electric vehicle controller
CN119690053A
