Intelligent Detection Method and System for Mixing Plants Based on Data Analysis

Abstract

This invention relates to the field of detection and training technology, specifically to an intelligent detection method and system for mixing plants based on data analysis. This invention combines the vibration signal characteristics of multiple parts of the mixing plant's mixer, decomposes and segments the signal components, adjusts the convergence trend of each segment, and comprehensively analyzes the convergence of similar vibration patterns. It also combines local convergence changes in working batches and abnormal fluctuations during local idling periods to obtain fault factors. By adjusting the model training for working batches using these fault factors, a more reliable training model is obtained. Furthermore, this invention analyzes abnormal characteristics by identifying when each historical working frequency exhibits more significant features during the idling periods between multiple adjacent mixing batches, thereby optimizing the signal weights in the prediction model construction and improving the detection results for mixing plants.

Description

Technical Field

[0001] This invention relates to the field of testing and training technology, and specifically to an intelligent testing method and system for mixing plants based on data analysis. Background Technology

[0002] As the "core hub" of concrete engineering, the concrete mixing plant is a crucial site for achieving large-scale, automated, and centralized concrete production, playing a decisive role in the quality control of concrete projects. Its equipped machinery, such as mixers, batching machines, and conveying systems, directly determines the accuracy of the concrete raw material proportions, the uniformity of mixing, and the quality of the finished product, thus affecting the overall construction quality and safety of the project. In actual operation, the mixing plant site experiences frequent malfunctions due to dust, large humidity fluctuations, and the equipment operating under continuous high-intensity conditions. These problems not only reduce production efficiency but can also lead to substandard concrete strength, creating serious hidden dangers to project quality.

[0003] When using predictive models to monitor the condition of mixers in a batching plant, the core strategy is to identify linear features in sensor data for prediction. However, the linear features of mixer vibration signals are relatively weak, especially the signal features reflecting potential abnormal faults are not obvious. As a result, in conventional predictive models, the abnormal signals corresponding to faults have a weak impact on the prediction results, making it difficult to accurately capture equipment anomalies. This can easily lead to missed or false diagnoses, and the inability to detect early faults such as bearing wear and belt conveyor misalignment in a timely manner. Consequently, production efficiency is affected, and potential engineering quality problems are created. Summary of the Invention

[0004] To address the technical problem in existing technologies where vibration signals and other linear characteristics are weak, resulting in unclear signal features reflecting potential abnormal faults, and consequently, weak influence of abnormal signals corresponding to faults on prediction results in conventional prediction models, making it difficult to accurately capture equipment anomalies and prone to missed or false detections, this invention aims to provide a data analysis-based intelligent detection method and system for mixing plants. The specific technical solution adopted is as follows:

[0005] This invention provides a data analysis-based intelligent detection method for mixing plants, the method comprising:

[0006] For each equipment component of the mixing plant, vibration signals and output current signals during each historical working batch and idle period are acquired; the idle period is between every two working batches; each vibration signal is decomposed into components, and segments are obtained by dividing the components and vibration signals into segments.

[0007] In each component, the attention weight of each segment is analyzed based on the deviation of each segment from other segments in terms of vibration intensity. Based on the attention weight of the segments, the similarity analysis between each component and the corresponding vibration signal in the peak fluctuation distribution is adjusted to obtain the convergence factor of each component.

[0008] For a single working batch, the fault factor of the working batch is obtained by the change of the convergence factor between adjacent working batches in time sequence and the degree of fluctuation of the output current signal during adjacent idling periods.

[0009] An anomaly detection model is trained based on the fault performance factors of each working batch, and the trained model is used for fault identification.

[0010] Furthermore, the method for obtaining the attention weight includes:

[0011] For any component, each segment of the component is taken as the target segment in turn, and the vibration intensity in the target segment is obtained.

[0012] The sum of the vibration intensities at the same location as the target segment on all components of the vibration signal corresponding to this component is taken as the total intensity of the target segment; the ratio of the vibration intensity of the target segment to the total intensity is taken as the intensity percentage of the target segment.

[0013] In this component, after calculating the difference in intensity ratio between the target segment and each other segment, the mean of all differences is normalized to obtain the attention weight of the target segment.

[0014] Furthermore, the method for obtaining the convergence factor includes:

[0015] Calculate the standard deviation of all peak values ​​in each segment to obtain the peak fluctuation index for each segment;

[0016] For any component, the peak fluctuation indexes are arranged according to the segmented distribution order of the component to obtain the peak trend sequence of the component; on the vibration signal corresponding to the component, the peak fluctuation indexes are arranged according to the segmented distribution order to obtain the peak trend sequence of the vibration signal.

[0017] By weighting the components according to their segmented attention weights, the similarity between the peak trend sequence of the component and the corresponding vibration signal is analyzed, and the convergence factor of the component is obtained.

[0018] Furthermore, the step of weighting the component with segmented attention weights and analyzing the similarity of the peak trends between the component and the corresponding vibration signal to obtain the convergence factor of the component includes:

[0019] When matching the peak trend sequence with the corresponding vibration signal using the DTW algorithm and calculating the shortest distance, the average attention weight between the segments corresponding to the peak fluctuation index in each matching result is used as the weight of each matching result; the distances between each matching result are weighted, summed, and negatively correlated to obtain the convergence factor of the component.

[0020] Furthermore, the method for obtaining the fault factor includes:

[0021] For any given work batch, sort the components of the work batch and the adjacent idle periods in the decomposition order to obtain the component sequence; based on the degree of convergence factor deviation between the components in the work batch and the components of the adjacent work batches in the time series, and the degree of correlation between the output current signal of the adjacent idle periods in the time series and the corresponding components in the component sequence, obtain the characterization influence of each component in the work batch.

[0022] By combining the characterization influence and intensity deviation of the components in two adjacent idle periods of the working batch, the intensity deviation of the working batch is obtained.

[0023] Analyze the degree of deviation of the output current signal fluctuation between two adjacent idle periods of the working batch to obtain the fluctuation coefficient deviation of the working batch;

[0024] By combining the fluctuation coefficient deviation and energy fluctuation deviation of this working batch, the failure factor of this working batch is obtained.

[0025] Furthermore, the method for obtaining the characterization influence includes:

[0026] The preceding work batch in terms of time sequence is designated as the previous batch, and the preceding idle period in terms of time sequence is designated as the previous idle period; the following work batch in terms of time sequence is designated as the next batch, and the following idle period in terms of time sequence is designated as the next idle period.

[0027] For any component in the working batch, take the component with the same component number in the previous batch as the previous target component, calculate the convergence factor difference between the previous target component and the component, and obtain the previous convergence deviation of the component.

[0028] In the component sequence of the pre-idle period, the component with the same sequence number as the component in the working batch is taken as the pre-analysis component; the correlation between the pre-analysis component and the output current signal of the pre-idle period is calculated to obtain the pre-correlation coefficient of the component; the product of the pre-correlation coefficient of the component and the pre-convergence deviation is taken as the pre-characterization degree of the component.

[0029] The component with the same sequence number as the component in the subsequent batch is taken as the subsequent target component. The convergence factor difference between the component and the subsequent target component is calculated to obtain the subsequent convergence deviation of the component.

[0030] In the component sequence after the idling period, the component with the same sequence number as the component in the working batch is taken as the post-analysis component; the correlation between the post-analysis component and the output current signal after the idling period is calculated to obtain the post-correlation coefficient of the component; the product of the post-correlation coefficient of the component and the post-convergence deviation is taken as the post-characterization degree of the component.

[0031] The mean values ​​of the post-characterization degree and the pre-characterization degree of the component are normalized to obtain the characterization influence degree of the component.

[0032] Furthermore, the method for obtaining the intensity deviation includes:

[0033] Obtain the vibration intensity of each component during the idling period; take the ratio of the vibration intensity of each component to the sum of the vibration intensities of all components on the corresponding vibration signal as the intensity ratio of each component; calculate the difference in intensity ratios of components with the same index in the component sequence between two adjacent idling periods of this working batch to obtain the deviation of each index.

[0034] After calculating the product of the representational influence and the deviation of each component corresponding to each index in the component sequence of the working batch, the mean of the products corresponding to all indexes is taken as the intensity deviation of the working batch.

[0035] Furthermore, the method for obtaining the fluctuation coefficient deviation includes:

[0036] Calculate the coefficient of variation of the output current signal for each adjacent idling period of the working batch, and use it as the fluctuation coefficient for each adjacent idling period;

[0037] The difference between the volatility coefficient of the latter idle period and the volatility coefficient of the former idle period in the time series of two adjacent idle periods of the work batch is taken as the volatility coefficient deviation of the work batch.

[0038] Furthermore, the method for obtaining the training model includes:

[0039] In the process of training the neural network model, the fault performance factor of each working batch is used as the weight of the loss function to train the vibration signal data of the working batch to obtain the training model.

[0040] The present invention also provides a data analysis-based intelligent detection system for a mixing plant, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the data analysis-based intelligent detection method for a mixing plant as described above.

[0041] The present invention has the following beneficial effects:

[0042] This invention combines the vibration signal characteristics of multiple parts of the mixer in a mixing plant, decomposes the signal components and segments them, and focuses on the convergence trend of each segment. It comprehensively analyzes the convergence of similar vibration patterns, considering the vibration characteristics as the mixing process moves from an unstable to a uniform and stable state, and obtains a convergence factor representing the single vibration pattern. Since wear and breakage become more pronounced with increasing batch numbers, the invention combines local convergence changes in working batches and abnormal fluctuations during local idling periods to comprehensively evaluate the salience of faults in a single working batch, obtaining a fault factor. The evaluated fault factor assigns different weights to each working batch for training the detection model, resulting in a more reliable training model. This invention analyzes abnormal characteristics by noting the more significant features of each historical working frequency during the idling period between multiple adjacent mixing batches, thereby optimizing the signal weights in the prediction model construction and improving the detection results for the mixing plant. Attached Figure Description

[0043] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart illustrating a data analysis-based intelligent detection method for a mixing plant, provided as an embodiment of the present invention;

[0045] Figure 2 This is a schematic diagram of the timing-based batch and idle distribution provided in one embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of the distribution of component segments provided in one embodiment of the present invention;

[0047] Figure 4 This is a flowchart of a method for obtaining fault factors according to an embodiment of the present invention. Detailed Implementation

[0048] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a data analysis-based intelligent detection method and system for mixing plants proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0050] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent detection method and system for mixing plants based on data analysis provided by this invention.

[0051] Please see Figure 1 The diagram illustrates a flowchart of an intelligent detection method for a mixing plant based on data analysis, according to an embodiment of the present invention. The method includes the following steps:

[0052] S1: For each equipment component of the mixing plant, acquire the vibration signal and the output current signal during each historical working batch and idle period; the idle period is between every two working batches; decompose each vibration signal into components, and divide the components and vibration signals into segments.

[0053] Vibration sensors are installed at key parts of the core equipment of the mixing plant to monitor different components. For example, in this embodiment of the invention, when testing the mixer, vibration signal acquisition covering radial, axial and tangential directions is deployed at the bearing housing, gearbox, mixing shaft coupling, mixing blade support and other locations. The sampling frequency is 10kHz. The implementer can adjust the sampling frequency according to the specific implementation scenario to ensure the capture of high-frequency fault characteristics, such as the impact signal of bearing ball breakage.

[0054] During routine operation at a batching plant, the concrete mix proportions to be mixed are typically determined in advance. Based on demand and the plant's maximum capacity, the mix is ​​prepared in multiple batches. Each batch is usually mixed for 3-5 minutes before being discharged into trucks to prepare for the next batch, which is then transported to the mixer. Therefore, vibration signals can be collected during each batch mixing process to obtain the vibration signal for each working batch. The period between the completion of one working batch and the addition of the next batch is called the idling period, which occurs between every two working batches. Please refer to [link to relevant documentation]. Figure 2 This illustrates a schematic diagram of a time-series batch and idle distribution provided by an embodiment of the present invention.

[0055] The data of a mixer when mixing stone and when running idle can vary greatly. Although the blades are rotating, the current power and vibration are very low when there is no material inside, which is normal. However, when there is an abnormal fault in the equipment, more obvious abnormalities will appear when running idle between batches. These abnormalities are more significant than when there is material inside.

[0056] To make the anomaly sensitivity of the subsequent training model more reliable, continuous historical working batches and idling periods are obtained in time, and vibration signals of each component are collected. Considering the more significant characteristics represented by idling period data, a Hall current sensor is further connected to the motor circuit to monitor the corresponding output current signal. The acquisition status is consistent with the vibration data. In this embodiment, a synchronous acquisition module is used to realize synchronous sampling of multi-sensor data with timestamp accuracy at the microsecond level to ensure that the data timing of vibration signal and output current signal is aligned.

[0057] The acquired vibration signal contains several components, such as the periodic movement of the machine, the collision of vibrating stones inside the mixer, and even the regular vibrations emitted by connecting parts during machine operation. As the stones in the mixer become more uniform, the resulting vibrations become more regular and smaller, and the vibration characteristics at each frequency gradually stabilize. When an abnormal fault occurs, irregular frequency signals will typically appear, and correspondingly, significant synchronous changes will be observed in multi-dimensional information such as current.

[0058] Therefore, to analyze abnormal situations more precisely, each vibration signal is decomposed into components and simultaneously segmented. In this embodiment of the invention, EMD decomposition can be used to obtain multiple IMF component signals corresponding to each vibration signal. After obtaining the period length corresponding to each component in the vibration signal, the least common multiple of the periods of the IMF components is calculated. Using the least common multiple as the segment length, each component and its corresponding vibration signal are segmented in the same way, ensuring that the number of periods for each component is an integer. Please refer to [link to relevant documentation]. Figure 3 The diagram illustrates a component segmentation distribution according to an embodiment of the present invention.

[0059] In this embodiment of the invention, the period of the component can be obtained by Hilbert transform. It should be noted that the methods of signal decomposition and period acquisition are well known to those skilled in the art. Signal decomposition can also be done by wavelet decomposition, etc., and period acquisition can also be done by autocorrelation function, etc., which will not be elaborated or limited here.

[0060] S2: In each component, analyze the attention weight of each segment based on the deviation of each segment from other segments in terms of vibration intensity; based on the attention weight of the segments, adjust the similarity analysis between each component and the corresponding vibration signal in the peak fluctuation distribution to obtain the convergence factor of each component.

[0061] In order to focus on analyzing the abnormalities in the composition, considering that as the mixing time increases, the ingredients in the same batch gradually become more uniform, and the resulting equipment vibration also tends to stabilize, while equipment failures, cracks and defects usually cause irregular vibrations, the analysis in segments can pay more attention to the parts with higher deviations in vibration intensity.

[0062] Preferably, in this embodiment of the invention, the method for obtaining the attention weight includes:

[0063] For any component, each segment of the component is taken as the target segment in turn, and each target segment is analyzed in turn to obtain the vibration intensity in the target segment. It should be noted that the vibration intensity is the root mean square of the signal in each segment. The acquisition of vibration intensity is a technique well known to those skilled in the art, and will not be elaborated here.

[0064] To ensure more reliable anomaly monitoring in component segments, the degree of inconsistency is analyzed by combining the proportion of vibration intensity of the segment across all components. First, the sum of the vibration intensities of all components corresponding to the target segment at the same location is taken as the total intensity of the target segment, reflecting the overall intensity at the target segment location. Then, the ratio of the vibration intensity of the target segment to the total intensity is taken as the intensity proportion of the target segment, indicating the proportion of that component within the target segment.

[0065] Furthermore, within this component, after calculating the difference in intensity proportion between the target segment and each other segment, the mean of all differences is normalized to obtain the attention weight of the target segment. The difference in intensity proportion between the target segment and other segments in the same component is calculated separately, and the mean of the difference in intensity proportion between the target segment and all other segments is calculated. In order to reflect the inconsistency of attention to different segments in the component, the mean is normalized to obtain the attention weight, making the subsequent performance analysis more accurate.

[0066] It should be noted that the normalization method can be linear normalization or standard normalization. Normalization is a well-known technique in the art, such as using the hyperbolic tangent function or the maximum-minimum normalization method, etc., and will not be specifically limited or elaborated here.

[0067] As the mixing time increases, the ingredients gradually blend evenly, and the equipment vibration tends to stabilize. Therefore, the fluctuations in each segment will also be a gradual convergence process. The peak fluctuation of a component reflects the vibration stability of the equipment at a certain frequency. However, faults and other anomalies do not exhibit this characteristic. When a component has a low similarity to the original signal, by analyzing its specific vibration, the proportion of fault information in each component can be identified to some extent.

[0068] Preferably, in this embodiment of the invention, the method for obtaining the convergence factor includes:

[0069] First, the standard deviation of all peak values ​​in each segment is calculated to obtain the peak fluctuation index for each segment. After extracting the peak values ​​in each segment, the standard deviation is calculated to reflect the degree of peak fluctuation in the segment, characterizing the vibration stability. It is understood that the acquisition of peak values ​​and the calculation of standard deviations are well-known techniques in the art and will not be elaborated here.

[0070] Furthermore, for any given component, the peak fluctuation indices are arranged according to the distribution order of the segments within that component to obtain the peak trend sequence of that component. Each segment of the component corresponds to a peak fluctuation index, and these indices are sorted by position to obtain a sequence reflecting the distribution of the peak fluctuation indices. Similarly, for the vibration signal acquisition sequence corresponding to that component, the peak fluctuation indices are arranged according to the distribution order of the segments within the vibration signal to obtain the peak trend sequence of the vibration signal. This sequence reflects the temporal stability changes of the vibration.

[0071] Finally, the similarity between the component and the corresponding vibration signal's peak trend sequence is analyzed by weighting the segmented attention weights, thus obtaining the component's convergence factor. In this embodiment, the DTW algorithm is used to more accurately quantify the similarity between sequences. When matching the peak trend sequence with the component and the corresponding vibration signal using the DTW algorithm and calculating the shortest distance, data point matching is performed first, which is equivalent to matching the peak fluctuation index in the two peak trend sequences. Then, the average attention weight between the segments corresponding to the peak fluctuation index in each matching result is used as the weight of each matching result. The comprehensive analysis of all attention weights in the matching results represents the proportion of the matching result in subsequent calculations. The smaller the shortest distance calculation result, the higher the trend similarity between sequences. Therefore, the distances between each matching result are weighted, summed, and negatively correlated to obtain the component's convergence factor.

[0072] It should be noted that the DTW algorithm and negative correlation mapping are techniques well known to those skilled in the art. Negative correlation mapping can be implemented in an inverse proportional form or a negative exponent form with the natural constant as the base, etc., which will not be elaborated here.

[0073] S3: For a single working batch, the fault factor of the working batch is obtained by the change of the convergence factor between adjacent working batches in the time sequence and the degree of fluctuation of the output current signal during adjacent idling periods.

[0074] The smaller the convergence factor, the more likely there are faults or defects in the current component structure. For concrete mixers, due to their long continuous working cycle, the initial local defects, such as wear or breakage of a single bearing ball, often become more serious as the number of batches increases. The proportion of the signal in the original signal also gradually increases, which will affect the accuracy of the component decomposition and cause fuzzy anomalies.

[0075] During the mixing process, the greatest vibration is caused by the ingredients being lifted and dropped by the mixing blades. Therefore, the equipment itself vibrates less during idling, while vibrations caused by equipment failure are relatively stronger. Thus, by combining the performance changes between adjacent batches and between adjacent idling periods, we can analyze the severity and increase of failures in each working batch, making the batch data more accurate for subsequent training.

[0076] Understandably, work batches that do not have two adjacent idle periods, namely the first work batch and the last work batch, are not included in the failure factor analysis, so as to make the data used for training more accurate and consistent, and the subsequent training model more reliable.

[0077] Preferably, in this embodiment of the invention, the method for obtaining the fault factor is described in [reference needed]. Figure 4 The diagram illustrates a flowchart of a method for obtaining fault factors according to an embodiment of the present invention, which includes the following steps:

[0078] S301: For any working batch, sort the components of the working batch and the adjacent idle period in the decomposition order to obtain the component sequence; based on the degree of convergence factor deviation between the components in the working batch and the components of the adjacent working batch in the time series, and the degree of correlation between the output current signal of the adjacent idle period in the time series and the corresponding component in the component sequence, obtain the characterization influence of each component in the working batch.

[0079] In existing signal decomposition, a fixed decomposition order is often used. Therefore, the components are sorted according to the decomposition order to facilitate the analysis of components at the same frequency level. Thus, the components in the working batch and the idle period are decomposed and sorted separately.

[0080] Adjacent batches and adjacent idle periods of a work batch in the time sequence are marked to facilitate subsequent analysis. In this embodiment of the invention, the work batch preceding the work batch in the time sequence is designated as the previous batch, the work batch preceding the work batch in the time sequence as the previous idle period in the time sequence is designated as the previous idle period in the time sequence, the work batch following the work batch in the time sequence is designated as the next batch in the time sequence, and the work batch following the work batch in the time sequence as the next idle period in the time sequence is designated as the next idle period in the time sequence.

[0081] For any component in the working batch, the component with the same sequence number as the component in the previous batch is taken as the previous target component. The convergence factor difference between the previous target component and the component is calculated to obtain the previous convergence deviation of the component. The components with the same sequence number in the previous batch are marked. By observing the convergence factor difference between components with the same frequency in the previous batch, the severity of the characteristic change can be reflected. The larger the convergence factor difference, the smaller the convergence factor of the component is compared with the previous batch, and the more significant the trend of severe fault change.

[0082] In the component sequence of the pre-idle period, the component with the same sequence number as the component in the working batch is taken as the pre-analysis component. The correlation between the pre-analysis component and the output current signal of the pre-idle period is calculated to obtain the pre-correlation coefficient of the component. The reliability of the abnormal characteristics is further reflected by the correlation consistency of the current signal during the idle period. The greater the correlation, the more reliable the analysis of the abnormal change of the component. In this embodiment of the invention, the correlation is calculated using the Pearson correlation coefficient. The use of the Pearson correlation coefficient to calculate the correlation is a well-known technique to those skilled in the art. Spearman's rank correlation coefficient, etc., can also be used, and no limitation or elaboration is made here.

[0083] By improving the sensitivity of anomaly detection during the idle period, the product of the pre-correlation coefficient and the pre-convergence deviation of the component is used as the pre-characteristic degree of the component, reflecting the degree of severity of the component's characterization compared to the preceding continuous case.

[0084] The same theory is used to judge the component and subsequent batches. The component with the same sequence number as the component in the subsequent batch is taken as the target component. The convergence factor difference between the component and the target component is calculated to obtain the convergence deviation of the component. The larger the convergence factor difference between the component and the target component, the smaller the convergence factor of the target component, and the more significant the trend of serious fault changes.

[0085] Similarly, in the component sequence after the idling period, the component with the same sequence number as the component in the working batch is taken as the post-analysis component. The correlation between the post-analysis component and the output current signal after the idling period is calculated to obtain the post-correlation coefficient of the component. The product of the post-correlation coefficient of the component and the post-convergence deviation is taken as the post-characterization degree of the component, reflecting the degree of severity of the component's characterization in the subsequent continuous case.

[0086] Finally, by combining the analysis before and after, the mean of the post-characterization degree and the pre-characterization degree of the component is normalized to obtain the characterization influence degree of the component, which represents the severity of the component's performance in consecutive adjacent batches.

[0087] S302: By combining the characterization influence and intensity deviation of the components in two adjacent idle periods of the working batch, the intensity deviation of the working batch is obtained.

[0088] By combining the intensity difference of the components during the idling period, the degree of stability failure is reflected, and by combining the characterization of the influence, the degree of anomaly with severe growth performance under vibration intensity analysis is obtained.

[0089] In this embodiment of the invention, the vibration intensity of each component during the idling period is obtained, and the method for obtaining the vibration intensity is the same as in step S2. The ratio of the vibration intensity of each component to the sum of the vibration intensities of all components on the corresponding vibration signal is taken as the intensity ratio of each component. The difference in intensity ratios of components with the same index in the component sequence between two adjacent idling periods of the working batch is calculated to obtain the deviation of each index. The degree of intensity change is reflected by the difference in proportion, which characterizes the occurrence of abnormal fluctuations.

[0090] Finally, by combining the characterization impact, the product of the characterization impact and the deviation of each component corresponding to each number in the component sequence of the work batch is calculated. The mean of the products corresponding to all numbers is taken as the intensity deviation of the work batch. By combining the severity of all components and the degree of abnormal fluctuation, the possible fault situations from continuous time series analysis are obtained. The greater the intensity deviation, the higher the probability of fault.

[0091] S303: Analyze the degree of deviation of the output current signal fluctuation between two adjacent idling periods of this working batch, and obtain the fluctuation coefficient deviation of this working batch.

[0092] When the difference between adjacent idling current data is large, the increase in the severity of faults in that batch may be more obvious. Therefore, in this embodiment of the invention, the coefficient of variation of the output current signal of each adjacent idling period of the working batch is calculated as the fluctuation coefficient of each adjacent idling period. It should be noted that the coefficient of variation is a technical means well known to those skilled in the art. The coefficient of variation is the ratio of the standard deviation to the mean, which is intended to reflect the degree of dispersion of the data. It will not be elaborated here.

[0093] Finally, the difference between the fluctuation coefficient of the latter idle period and the fluctuation coefficient of the former idle period in the timing sequence of two adjacent idle periods of the working batch is taken as the fluctuation coefficient deviation of the working batch, which reflects the greater deviation in the output current fluctuation. The larger the fluctuation coefficient deviation, the more serious the fault performance.

[0094] S304: Combine the fluctuation coefficient deviation and energy fluctuation deviation of this working batch to obtain the failure factor of this working batch.

[0095] Finally, by combining the vibration intensity analysis of the batch and the current fluctuation analysis during the idling period, a fault factor that more accurately characterizes the degree of abnormality is obtained. In this embodiment of the invention, the product of the fluctuation coefficient deviation and the energy fluctuation deviation of the working batch is used as the fault factor of the working batch. The larger the fault factor, the higher the weight of the working batch in fault identification and detection.

[0096] S4: Train an anomaly detection model based on the fault performance factors of each working batch, and use the trained model for fault identification.

[0097] Model training can be performed based on fault factors from historical work batches, adjusting the emphasis on identifying signal anomalies to improve accuracy. The fault performance factor reflects the severity of fault anomalies in each work batch; a higher fault performance factor indicates more important data in that batch and greater value for model training.

[0098] In this embodiment of the invention, during the neural network training model, the fault performance factor of each working batch is used as the weight of the loss function to train the vibration signal data of the working batches to obtain the training model. By using the fault performance factor as the weight and modifying the calculation method of the loss function, that is, multiplying the loss of each working batch by the corresponding fault performance factor when calculating the loss, batches with higher fault performance factors can have a greater impact on the updating of model parameters during optimization. It should be noted that the neural network training model is a technique well known to those skilled in the art and will not be described in detail here.

[0099] Based on the trained model, fault identification can be performed on the vibration data of various components in the mixing plant collected in real time. The fault identification results can then be used to establish an early warning mechanism. In this embodiment of the invention, for example, when the fault probability output by the model of a certain part of a single device is greater than 0.6, a Level 1 warning can be issued, the anomaly recorded, and a notification sent to the on-duty personnel to prompt them to strengthen monitoring. When more than two parts of the same device issue warnings simultaneously, or when the difference in fault probabilities detected consecutively exceeds 0.3, a Level 2 warning can be issued, automatically switching to a high-frequency sampling mode and activating the fault tracing algorithm. When different devices issue warnings in a chain reaction or the detected fault probability exceeds 0.9, a Level 3 warning can be issued, triggering an emergency shutdown procedure, locking the equipment operating interface, and notifying the maintenance team.

[0100] In summary, this invention combines the vibration signal characteristics of multiple parts of the mixing plant's mixer, decomposes and segments the signal components, adjusts the convergence trend of each segment, comprehensively analyzes the convergence of similar vibration patterns, and considers the vibration characteristics of the mixing process as it moves from an unstable to a uniform and stable state. This yields a convergence factor characterizing the single vibration pattern. Since wear and breakage become more pronounced with increasing batch numbers, the invention combines local convergence changes in working batches with abnormal fluctuations during local idling periods to comprehensively evaluate the salience of faults in a single working batch, resulting in a fault factor. The evaluated fault factor assigns different weights to each working batch for training the detection model, leading to a more reliable training model. Furthermore, this invention analyzes abnormal characteristics by noting the more significant features of each historical working frequency during the idling periods between multiple adjacent mixing batches, thereby optimizing the signal weights in the prediction model construction and improving the detection results for the mixing plant.

[0101] The present invention also provides a data analysis-based intelligent detection system for a mixing plant, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the data analysis-based intelligent detection method for a mixing plant as described above.

[0102] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0103] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A data analysis-based intelligent detection method for mixing plants, characterized in that, The method includes: For each equipment component of the mixing plant, vibration signals and output current signals during each historical working batch and idle period are acquired; the idle period is between every two working batches; each vibration signal is decomposed into components, and segments are obtained by dividing the components and vibration signals into segments. In each component, the attention weight of each segment is analyzed based on the deviation of each segment from other segments in terms of vibration intensity. Based on the attention weight of the segments, the similarity analysis between each component and the corresponding vibration signal in the peak fluctuation distribution is adjusted to obtain the convergence factor of each component. For a single working batch, the fault factor of the working batch is obtained by the change of the convergence factor between adjacent working batches in time sequence and the degree of fluctuation of the output current signal during adjacent idling periods. An anomaly detection model is trained based on the fault performance factors of each work batch, and the trained model is used for fault identification. The methods for obtaining the attention weights include: For any component, each segment of the component is taken as the target segment in turn, and the vibration intensity in the target segment is obtained. The sum of the vibration intensities at the same location as the target segment on all components of the vibration signal corresponding to this component is taken as the total intensity of the target segment; the ratio of the vibration intensity of the target segment to the total intensity is taken as the intensity percentage of the target segment. In this component, after calculating the difference in intensity ratio between the target segment and each other segment, the mean of all differences is normalized to obtain the attention weight of the target segment. The method for obtaining the convergence factor includes: Calculate the standard deviation of all peak values ​​in each segment to obtain the peak fluctuation index for each segment; For any component, the peak fluctuation indexes are arranged according to the segmented distribution order of the component to obtain the peak trend sequence of the component; on the vibration signal corresponding to the component, the peak fluctuation indexes are arranged according to the segmented distribution order to obtain the peak trend sequence of the vibration signal. By weighting the components according to their segmented attention weights, the similarity between the peak trend sequence of the component and the corresponding vibration signal is analyzed, and the convergence factor of the component is obtained. The method for obtaining the fault factor includes: For any given work batch, sort the components of the work batch and the adjacent idle periods in the decomposition order to obtain the component sequence; based on the degree of convergence factor deviation between the components in the work batch and the components of the adjacent work batches in the time series, and the degree of correlation between the output current signal of the adjacent idle periods in the time series and the corresponding components in the component sequence, obtain the characterization influence of each component in the work batch. By combining the characterization influence and intensity deviation of the components in two adjacent idle periods of the working batch, the intensity deviation of the working batch is obtained. Analyze the degree of deviation of the output current signal fluctuation between two adjacent idle periods of the working batch to obtain the fluctuation coefficient deviation of the working batch; By combining the fluctuation coefficient deviation and energy fluctuation deviation of this working batch, the failure factor of this working batch is obtained.

2. The intelligent detection method for a mixing plant based on data analysis according to claim 1, characterized in that, The method of weighting by segmented attention weights, analyzing the similarity of the peak trend between the component and the corresponding vibration signal, and obtaining the convergence factor of the component includes: When matching the peak trend sequence with the corresponding vibration signal using the DTW algorithm and calculating the shortest distance, the average attention weight between the segments corresponding to the peak fluctuation index in each matching result is used as the weight of each matching result; the distances between each matching result are weighted, summed, and negatively correlated to obtain the convergence factor of the component.

3. The intelligent detection method for a mixing plant based on data analysis according to claim 1, characterized in that, The method for obtaining the characterization influence includes: The preceding work batch in terms of time sequence is designated as the previous batch, and the preceding idle period in terms of time sequence is designated as the previous idle period; the following work batch in terms of time sequence is designated as the next batch, and the following idle period in terms of time sequence is designated as the next idle period. For any component in the working batch, take the component with the same component number in the previous batch as the previous target component, calculate the convergence factor difference between the previous target component and the component, and obtain the previous convergence deviation of the component. In the component sequence of the pre-idle period, the component with the same sequence number as the component in the working batch is taken as the pre-analysis component; the correlation between the pre-analysis component and the output current signal of the pre-idle period is calculated to obtain the pre-correlation coefficient of the component; the product of the pre-correlation coefficient of the component and the pre-convergence deviation is taken as the pre-characterization degree of the component. The component with the same sequence number as the component in the subsequent batch is taken as the subsequent target component. The convergence factor difference between the component and the subsequent target component is calculated to obtain the subsequent convergence deviation of the component. In the component sequence after the idling period, the component with the same sequence number as the component in the working batch is taken as the post-analysis component; the correlation between the post-analysis component and the output current signal after the idling period is calculated to obtain the post-correlation coefficient of the component; the product of the post-correlation coefficient of the component and the post-convergence deviation is taken as the post-characterization degree of the component. The mean values ​​of the post-characterization degree and the pre-characterization degree of the component are normalized to obtain the characterization influence degree of the component.

4. The intelligent detection method for a mixing plant based on data analysis according to claim 1, characterized in that, The method for obtaining the strength deviation includes: Obtain the vibration intensity of each component during the idling period; take the ratio of the vibration intensity of each component to the sum of the vibration intensities of all components on the corresponding vibration signal as the intensity ratio of each component; calculate the difference in intensity ratios of components with the same index in the component sequence between two adjacent idling periods of this working batch to obtain the deviation of each index. After calculating the product of the representational influence and the deviation of each component corresponding to each index in the component sequence of the working batch, the mean of the products corresponding to all indexes is taken as the intensity deviation of the working batch.

5. The intelligent detection method for a mixing plant based on data analysis according to claim 1, characterized in that, The method for obtaining the fluctuation coefficient deviation includes: Calculate the coefficient of variation of the output current signal for each adjacent idling period of the working batch, and use it as the fluctuation coefficient for each adjacent idling period; The difference between the volatility coefficient of the latter idle period and the volatility coefficient of the former idle period in the time series of two adjacent idle periods of the work batch is taken as the volatility coefficient deviation of the work batch.

6. The intelligent detection method for a mixing plant based on data analysis according to claim 1, characterized in that, The method for obtaining the training model includes: In the process of training the neural network model, the fault performance factor of each working batch is used as the weight of the loss function to train the vibration signal data of the working batch to obtain the training model.

7. A data analysis-based intelligent detection system for a mixing plant, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent detection method for mixing plants based on data analysis as described in any one of claims 1 to 6.

Images