A multi-source data fusion industrial equipment health diagnosis method
By constructing operating condition labels and generating real-time curves through correlation analysis, the problem of needing to shut down the machine for shaft health diagnosis was solved, realizing real-time diagnosis and reducing operating costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 浙江恩赫控股集团有限公司
- Filing Date
- 2026-03-09
- Publication Date
- 2026-06-16
AI Technical Summary
In existing technologies, shaft health diagnosis requires equipment shutdown, which cannot achieve real-time diagnosis and increases operating costs.
By acquiring standards and fault curves, operating condition labels are constructed, fusion weights are generated based on correlation analysis, real-time multi-source data is reorganized to generate real-time curves, and weighted fusion is performed to generate health information.
It enables shaft health diagnosis without shutting down the machine, reducing equipment operating costs.
Smart Images
Figure CN122220752A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of industrial equipment health diagnosis, and in particular to a method for industrial equipment health diagnosis using multi-source data fusion. Background Technology
[0002] In industrial equipment such as elevators, large shipping equipment, steam turbines, and homogenizers, shafts are core components for power transmission, and their health directly affects the operational safety and production efficiency of the equipment. Once a shaft experiences wear, cracks, or bearing failure, it can lead to equipment downtime or even serious safety accidents. Therefore, timely and accurate health diagnosis of shafts is a crucial aspect of industrial equipment maintenance.
[0003] Currently, the industry's health diagnostic solutions for hinges generally follow the "preset program calibration - offline comparison and judgment" model. Specifically, the implementation process of this model is roughly divided into two stages:
[0004] The first stage is the generation of standard curves and fault curves. This involves driving the shaft through a preset control program to ensure stable operation under various preset working conditions (including different speed and torque combinations). Simultaneously, multi-source data of the shaft under normal conditions and multi-source data under fault conditions are collected through detection devices such as sound sensors and vibration sensors. These data are then organized into standard curves and fault curves according to the time series.
[0005] The second stage is health status determination, which means that after the equipment has been running for a period of time, the normal production process of the equipment is interrupted by stopping the machine, and the preset program is restarted to make the shaft run according to the working condition mode of the first stage. Multi-source data is collected again and real-time curves are generated. The real-time curves are compared with the pre-stored standard curves and fault curves, and the current health status of the shaft is determined according to the degree of matching of the curves.
[0006] This method of health diagnosis relies on a preset program running after the equipment is shut down, which not only fails to achieve real-time diagnosis but also significantly increases the operating cost of the equipment. Summary of the Invention
[0007] To address the aforementioned issues, this application provides a method for industrial equipment health diagnosis based on multi-source data fusion.
[0008] This application provides a method for industrial equipment health diagnosis through multi-source data fusion, which adopts the following technical solution:
[0009] A method for health diagnosis of industrial equipment using multi-source data fusion, comprising:
[0010] Obtain standard curves and fault curves, and set condition labels according to the testing conditions; both standard curves and fault curves include sound curves and vibration curves.
[0011] Based on the standard curve and the fault curve, a fusion weight is generated through correlation analysis. The fusion weight includes sound weight and vibration weight.
[0012] Acquire real-time multi-source data and bind the real-time multi-source data within a preset time window to operating condition labels; the real-time multi-source data includes real-time time, real-time speed, real-time torque, real-time sound, and real-time vibration;
[0013] Real-time multi-source data is reconstructed based on working condition labels to generate real-time curves. The real-time curves are compared with standard curves and fault curves to obtain sound judgment results and vibration judgment results, respectively.
[0014] Based on the fusion weight, the sound judgment result and the vibration judgment result are weighted and fused to generate health information.
[0015] In one embodiment: operating condition labels are constructed based on the speed, torque, and time at each point on the standard curve and the fault curve, and the operating condition labels are sorted by time.
[0016] In one embodiment, the correlation analysis method is a random forest algorithm, a gradient boosting tree algorithm, or a mutual information algorithm.
[0017] In one embodiment: based on real-time speed and real-time torque, real-time multi-source data is bound to operating condition labels using preset rules; wherein, the preset rules are within a preset range of speed and torque corresponding to the operating condition labels.
[0018] In one embodiment: the steps of reconstructing real-time multi-source data based on operating condition labels and generating real-time curves, comparing the real-time curves with standard curves and fault curves, and obtaining sound judgment results and vibration judgment results respectively include:
[0019] Real-time multi-source data with the same working condition label are sorted according to real-time time, and then real-time curves are constructed based on the working condition label order and real-time time order. The real-time curves include real-time sound curves and real-time vibration curves.
[0020] The real-time curves were fitted to the standard curve and the fault curve respectively to obtain the sound judgment results and vibration judgment results respectively.
[0021] In one embodiment: when constructing a real-time curve, if the real-time multi-source data in the working condition label is greater than the required data volume, it is reduced to the required data volume by a sliding window feature extraction method.
[0022] If there is no real-time multi-source data in the working condition label, or the real-time multi-source data is less than the required amount of data, the real-time multi-source data in the previous N preset time windows will be obtained sequentially until the required amount of data is obtained.
[0023] In one embodiment, the step of fitting the real-time curve to the standard curve and the fault curve respectively to obtain the sound judgment result and the vibration judgment result specifically includes:
[0024] Based on the operating condition labels, the real-time curve, standard curve, and fault curve are segmented, and the fitting parameters of each segment are obtained by segmented spline fitting.
[0025] By performing correlation analysis on the standard curve and the fault curve, the fault correlation coefficient corresponding to each operating condition label is obtained;
[0026] The fitted parameters and fault correlation coefficients are weighted and fused to obtain the sound judgment result and the vibration judgment result.
[0027] In one embodiment: the standard curve and the fault curve further include a temperature curve, the fusion weights further include a temperature weight, and the real-time multi-source data further includes real-time temperature; the health diagnosis method further includes:
[0028] The preset learning model is trained using temperature curves to obtain a judgment model;
[0029] The temperature determination result is obtained by taking real-time multi-source data within a preset time window as input and using a determination model.
[0030] Based on the fusion weights, the sound judgment results, vibration judgment results, and temperature judgment results are weighted and fused to generate health information.
[0031] In one embodiment: the training method of the judgment model is as follows:
[0032] Based on a preset duration, multiple temperature curve segments under the working condition label are obtained, and the temperature curve segments within each working condition label are divided into training set and validation set according to the proportion.
[0033] The running time is taken as the start time of the temperature curve segment;
[0034] The trend characteristics of temperature curve segments are used as feature vectors;
[0035] The model is trained by taking the working condition labels, feature vectors and running time in the training set as inputs and the health status labels as outputs, and a judgment model is obtained.
[0036] The decision model is validated by using the operating condition labels, feature vectors, and runtime in the validation set as inputs, and the decision model is optimized based on the validation results.
[0037] In summary, this application has the following beneficial effects: by binding operating condition tags, real-time data is reorganized to generate real-time curves, enabling the generation of health status data through real-time data without stopping the machine. Attached Figure Description
[0038] Figure 1 This is a flowchart of a multi-source data fusion method for industrial equipment health diagnosis in this embodiment;
[0039] Figure 2 This is a flowchart of another industrial equipment health diagnosis method based on multi-source data fusion in this embodiment. Detailed Implementation
[0040] The present application will be further described in detail below with reference to the accompanying drawings.
[0041] To better understand the purpose, technical solutions, and advantages of this application, it has been described and illustrated below with reference to the accompanying drawings and embodiments. However, those skilled in the art should understand that this application can be implemented without these details. In some cases, to avoid obscuring various aspects of this application due to unnecessary description, well-known methods, processes, systems, components, and / or circuits already described at a higher level will not be elaborated upon. It will be apparent to those skilled in the art that various modifications can be made to the embodiments disclosed in this application, and the general principles defined in this application can be applied to other embodiments and application scenarios without departing from the principles and scope of this application. Therefore, this application is not limited to the illustrated embodiments, but conforms to the broadest scope consistent with the scope of protection claimed in this application.
[0042] It should be noted that the descriptions of these embodiments are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0043] In the description of this application, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.
[0044] In the description of this application, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples.
[0045] A method for health diagnosis of industrial equipment using multi-source data fusion, such as Figure 1 As shown, it includes the following steps:
[0046] S100: Obtain the standard curve and fault curve, and set the working condition label according to the test conditions; wherein, both the standard curve and the fault curve include the sound curve and the vibration curve.
[0047] In this step, operating condition labels are constructed based on the speed, torque, and time corresponding to each point on the standard curve and the fault curve, using speed and torque as the data points. The time length of the operating condition labels does not need to be uniform during construction; it can be set according to the duration of each stage.
[0048] The process involves setting a stable phase where the speed and torque are fixed, and then further subdividing the stable phase into multiple phases based on the different speeds and torques. The phase where the speed changes is set as the variable phase. Each phase must be continuous in time and is marked with a unique operating condition label.
[0049] After completing the construction of the working condition labels, sort the working condition labels according to the time of each stage, which can be based on the start time, end time or intermediate time of the stage.
[0050] S200, based on standard curves and fault curves, generates fusion weights through correlation analysis. The fusion weights include sound weights and vibration weights.
[0051] In this step, the correlation analysis method can be random forest algorithm, gradient boosting tree algorithm or mutual information algorithm. Among them, since the mutual information algorithm has a lower requirement for the amount of working condition data, it is preferred to use the mutual information algorithm.
[0052] S300: Acquire real-time multi-source data and bind the real-time multi-source data within a preset time window to the operating condition label; wherein, the real-time multi-source data includes real-time time, real-time speed, real-time torque, real-time sound and real-time vibration.
[0053] In this step, based on the real-time speed and torque of the real-time multi-source data, the speed and torque are compared with those corresponding to the operating condition label. The real-time multi-source data are bound to the operating condition label using preset rules. The number of real-time multi-source data under each operating condition label can be arbitrary.
[0054] The preset rule is that the speed and torque corresponding to the operating condition label are within a preset range. By setting the range, the data within a certain fluctuation range still has high judgment accuracy. At the same time, this method can also bind more real-time multi-source data to the operating condition label.
[0055] S400: Based on the working condition label, real-time multi-source data is recombined and real-time curves are generated. The real-time curves are compared with the standard curves and fault curves to obtain sound judgment results and vibration judgment results respectively.
[0056] In this step, when generating the real-time curve, real-time multi-source data with the same operating condition label are sorted according to real-time time. Curves are then generated by sorting the real-time multi-source data under the operating condition label according to time. Finally, the generated curves are combined based on the time order of the operating condition labels to form a complete real-time curve. Specifically, sound and vibration data are used to construct independent curves; therefore, the real-time curve includes a real-time sound curve and a real-time vibration curve.
[0057] It should be noted that when constructing real-time curves, the real-time operating conditions are set according to the needs of the work, and their duration varies. Therefore, within the preset time window, the real-time multi-source data bound to the operating condition label may be zero, greater than the required data volume, or less than the required data volume. The required data volume refers to the number of data points within the operating condition label in the standard curve and the fault curve.
[0058] For example, in this embodiment, the above problems are solved by reducing the amount of data through data completion and smoothing processes.
[0059] Specifically, if the real-time multi-source data within the operating condition label exceeds the required data volume, it is reduced to the required data volume using a sliding window feature extraction method. That is, based on the difference between the real-time multi-source data and the required data volume, a fixed sliding window is designed. The sliding window contains at least two sets of real-time multi-source data. A new set of data is obtained by averaging the real-time multi-source data within the sliding window. Then, the sliding parameters of the sliding window are set according to the required data volume.
[0060] Taking two sets of real-time multi-source data within a sliding window as an example, the sliding parameter can be set to 1 or 2. When set to 1, the amount of new data is one less than the amount of original real-time multi-source data. When set to 2, the amount of new data is half the amount of original real-time multi-source data.
[0061] If there is no real-time multi-source data in the working condition label, or the real-time multi-source data is less than the required amount of data, the real-time multi-source data in the previous N preset time windows will be obtained sequentially until the required amount of data is obtained.
[0062] Regarding the comparison between the real-time curve and the standard curve and the fault curve, this embodiment uses curve fitting. The sound judgment result is obtained by comparing the real-time sound curve with the sound curves in the standard curve and the fault curve respectively; the vibration judgment result is obtained by comparing the real-time vibration curve with the vibration curves in the standard curve and the fault curve respectively.
[0063] The acquisition principles for sound judgment results and vibration judgment results are the same. This embodiment uses the acquisition principle of sound judgment results as an example. After comparing the real-time sound curve with the sound curves in the standard curve and the fault curve respectively, sound standard fit degree and sound fault fit degree will be generated respectively. The value range of sound standard fit degree and sound fault fit degree is 0-1. The higher the sound standard fit degree, the healthier the sound is, and the lower the sound fault fit degree, the healthier the sound is. Therefore, the sound judgment result = (sound standard fit degree + (1 - sound fault fit degree)) / 2.
[0064] The curve fitting method is either polynomial fitting or piecewise spline fitting. In this embodiment, piecewise spline fitting is preferred. Specifically, the curve fitting steps in this embodiment are as follows:
[0065] Based on the operating condition labels, the real-time curve, standard curve, and fault curve are segmented, and the fitting parameters of each segment are obtained by segmented spline fitting.
[0066] By performing correlation analysis on the standard curve and the fault curve, the fault correlation coefficient corresponding to each operating condition label is obtained;
[0067] The fitted parameters and fault correlation coefficients are weighted and fused to obtain the sound judgment result and the vibration judgment result.
[0068] In the above fitting method, segments are made by using working condition labels. At the same time, the influence of data on the results under different working conditions is obtained by correlation analysis. The principle of correlation analysis here is the same as that of correlation analysis in step S200.
[0069] S500: Based on fusion weights, the sound judgment results and vibration judgment results are weighted and fused to generate health information.
[0070] In one embodiment, since temperature is also an important indicator of abnormal conditions of the shaft in addition to sound and vibration, temperature data is introduced in this embodiment, and a comprehensive judgment is made by using three sets of data: temperature, sound, and vibration.
[0071] Correspondingly, the standard curve and fault curve also include the temperature curve, the fusion weight also includes the temperature weight, and the real-time multi-source data also includes the real-time temperature.
[0072] With the introduction of temperature parameters, the steps of the health diagnosis method also include:
[0073] S600: Train the preset learning model using the temperature curve to obtain the judgment model.
[0074] In this step, such as Figure 2 As shown, the training method for the judgment model is as follows:
[0075] S601. Based on a preset duration, acquire multiple temperature curve segments under the operating condition label, and divide the temperature curve segments within each operating condition label into a training set and a validation set according to the proportion.
[0076] S602, use the start time of the temperature curve segment as the running time.
[0077] S603. Use the trend characteristics of temperature curve segments as feature vectors.
[0078] S604. Using the working condition labels, feature vectors, and runtime from the training set as inputs and the health status labels as outputs, train the learning model to obtain the judgment model.
[0079] S605. Use the working condition labels, feature vectors and running time in the validation set as inputs to validate the decision model, and optimize the decision model based on the validation results.
[0080] By segmenting the curve within the operating condition label, the trend characteristics of the temperature curve segment are used as the feature vector. Statistical features include maximum value, minimum value, average value, standard deviation, etc. The trend feature refers to the overall slope, and the abrupt change feature refers to the number of sudden change points.
[0081] If the temperature curve segment originates from the standard curve, the health status is labeled as healthy; if it originates from the fault curve, the health status is labeled as faulty.
[0082] S700: Takes real-time multi-source data within a preset time window as input and obtains temperature determination results through a determination model.
[0083] In this step, before inputting the real-time multi-source data within the preset time window, a real-time temperature curve is pre-constructed. The real-time temperature curve is constructed directly based on the acquisition time of the real-time multi-source data.
[0084] Then, the operating condition label is bound by the speed and torque. After binding, the curve within the operating condition label is segmented according to the same preset duration in step S601. Finally, feature vectors and runtime data are generated as input to the model.
[0085] S800, based on fusion weights, performs weighted fusion of sound judgment results, vibration judgment results and temperature judgment results to generate health information.
[0086] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.
Claims
1. A method for health diagnosis of industrial equipment using multi-source data fusion, characterized in that, include: Obtain standard curves and fault curves, and set condition labels according to the testing conditions; both standard curves and fault curves include sound curves and vibration curves. Based on the standard curve and the fault curve, a fusion weight is generated through correlation analysis. The fusion weight includes sound weight and vibration weight. Acquire real-time multi-source data and bind the real-time multi-source data within a preset time window to operating condition labels; the real-time multi-source data includes real-time time, real-time speed, real-time torque, real-time sound, and real-time vibration; Real-time multi-source data is reconstructed based on working condition labels to generate real-time curves. The real-time curves are compared with standard curves and fault curves to obtain sound judgment results and vibration judgment results, respectively. Based on the fusion weight, the sound judgment result and the vibration judgment result are weighted and fused to generate health information.
2. The industrial equipment health diagnosis method based on multi-source data fusion according to claim 1, characterized in that: Based on the speed, torque, and time at each point on the standard curve and the fault curve, operating condition labels are constructed using speed and torque, and then sorted by time.
3. The industrial equipment health diagnosis method based on multi-source data fusion according to claim 1, characterized in that: The correlation analysis methods are random forest algorithm, gradient boosting tree algorithm or mutual information algorithm.
4. The industrial equipment health diagnosis method based on multi-source data fusion according to claim 1, characterized in that: Based on real-time speed and real-time torque, real-time multi-source data is bound to operating condition labels using preset rules; the preset rules are those that are within a preset range of speed and torque corresponding to the operating condition label.
5. The industrial equipment health diagnosis method based on multi-source data fusion according to claim 1, characterized in that, The specific steps for reconstructing real-time multi-source data based on operating condition labels and generating real-time curves, comparing the real-time curves with standard curves and fault curves to obtain sound judgment results and vibration judgment results respectively, include: Real-time multi-source data with the same working condition label are sorted according to real-time time, and then real-time curves are constructed based on the working condition label order and real-time time order. The real-time curves include real-time sound curves and real-time vibration curves. The real-time curves were fitted to the standard curve and the fault curve respectively to obtain the sound judgment results and the vibration judgment results respectively.
6. The industrial equipment health diagnosis method based on multi-source data fusion according to claim 5, characterized in that: When constructing real-time curves, if the real-time multi-source data in the working condition label is greater than the required data volume, the data volume is reduced to the required volume by using a sliding window feature extraction method. If there is no real-time multi-source data in the working condition label, or the real-time multi-source data is less than the required amount of data, the real-time multi-source data in the previous N preset time windows will be obtained sequentially until the required amount of data is obtained.
7. The industrial equipment health diagnosis method based on multi-source data fusion according to claim 5, characterized in that, The specific steps for fitting the real-time curve to the standard curve and the fault curve to obtain the sound judgment result and the vibration judgment result respectively include: Based on the operating condition labels, the real-time curve, standard curve, and fault curve are segmented, and the fitting parameters of each segment are obtained by segmented spline fitting. By performing correlation analysis on the standard curve and the fault curve, the fault correlation coefficient corresponding to each operating condition label is obtained; The fitted parameters and fault correlation coefficients are weighted and fused to obtain the sound judgment result and the vibration judgment result.
8. The industrial equipment health diagnosis method based on multi-source data fusion according to claim 1, characterized in that, The standard curve and fault curve also include a temperature curve, the fusion weights also include a temperature weight, and the real-time multi-source data also includes real-time temperature; the health diagnosis method also includes: The preset learning model is trained using temperature curves to obtain a judgment model; The temperature determination result is obtained by taking real-time multi-source data within a preset time window as input and using a determination model. Based on the fusion weights, the sound judgment results, vibration judgment results, and temperature judgment results are weighted and fused to generate health information.
9. The industrial equipment health diagnosis method based on multi-source data fusion according to claim 8, characterized in that, The training method for the judgment model is as follows: Based on a preset duration, multiple temperature curve segments under the working condition label are obtained, and the temperature curve segments within each working condition label are divided into training set and validation set according to the proportion. The running time is taken as the start time of the temperature curve segment; The trend characteristics of temperature curve segments are used as feature vectors; The model is trained by taking the working condition labels, feature vectors and running time in the training set as inputs and the health status labels as outputs, and a judgment model is obtained. The decision model is validated by using the operating condition labels, feature vectors, and runtime in the validation set as inputs, and the decision model is optimized based on the validation results.