CN-121996913-A - Method and system for predicting using state of crusher lining plate based on TFT

CN121996913ACN 121996913 ACN121996913 ACN 121996913ACN-121996913-A

Abstract

The invention aims to solve the problems of the existing breaker lining plate service life prediction technology and provides a breaker lining plate service state prediction method and system based on a TFT. In the present invention, TFTs are used as the core backbone structure of the model, and an improved module for explicit alignment and multitasking prediction is added on this basis. The method comprises the steps of obtaining data related to lining plate consumption, preprocessing the data to obtain time sequence data and scalar data, reconstructing a life prediction model, and carrying out feature extraction and prediction on the obtained data to obtain a prediction result. According to the method, a lining plate sensor is not needed, so that the cost of a method for predicting the service life of the lining plate is reduced, the prediction accuracy is improved, the state and the service condition of the lining plate of the crusher can be analyzed in real time, the abnormal shutdown condition of equipment caused by the abrasion of the lining plate is greatly reduced, and the production and stable production are improved.

Inventors

HU JIAN
LIU TAIHE
LIU CHANG
LIU XIAOWEI
ZHANG WENHUI
LI HONG
Gai Junpeng
XIAO CHENGYONG

Assignees

鞍钢集团关宝山矿业有限公司

Dates

Publication Date: 20260508
Application Date: 20260107

Claims (10)

1. The method for predicting the use state of the crusher lining plate based on the TFT is characterized by comprising the following steps of: S1, acquiring production data and lining plate technical data related to the consumption of a lining plate of a crusher, wherein the production data and the lining plate technical data comprise time series data consisting of historical data and scalar data which do not change with time, preprocessing the data to obtain a data set of the time series data and a data set of the scalar data; S2, constructing a life prediction model; The life prediction model is based on a TFT and comprises a feature extraction module and a prediction module, wherein the feature extraction module comprises a time sequence processing unit, a scalar processing unit and a multi-head attention unit, and the prediction module consists of an LSTM unit and a multi-task prediction unit comprising a plurality of parallel FC layers, wherein the plurality of parallel FC layers are respectively connected with the LSTM unit; S21, extracting the characteristics of the data; Extracting data features from the data through a time sequence processing unit and a scalar processing unit of the model to respectively obtain time sequence features and scalar features, fusing the time sequence features and the scalar features, and obtaining feature vectors finally output by a feature extraction module through a multi-head attention unit; S22, inputting the final feature vector of the feature extraction module into a prediction module of the model, and predicting three tasks of a lining plate abrasion state, a crusher running state and a lining plate residual service life RUL by adopting a multi-task learning frame of a multi-task prediction unit; s3, training a life prediction model; S4, collecting time sequence data and scalar data of the crusher lining plate in actual production, and predicting the using state of the lining plate by using a trained model.
2. The TFT-based crusher lining plate usage state prediction method according to claim 1, wherein in S1, the time series data includes ore hardness, grain size composition, crusher power, bin level, ore feeding amount, ore feeding grain size, ore discharging grain size, and lining plate replacement cycle; The scalar data comprises the shape, structure, material, hardness, model number and ore breaking section of the lining plate.
3. The TFT-based crusher lining plate use state prediction method of claim 1, wherein in S1, the method of preprocessing data is as follows: s11, removing irrelevant data and abnormal data; S12, integrating the same data source data, and unifying time stamps to form a unified data set; Integrating the data source data comprises extracting common attribute columns from different data sources, unifying time stamps, synchronizing and interpolating the time stamps of the multi-source data by an explicit time alignment method, so that the data of different sampling frequencies and time windows are kept consistent in the time dimension; S13, performing minimum-maximum normalization on the data in the data set, and scaling the data to be within a [0,1] interval; The conversion is performed using the following formula: ; Where x represents the raw data point, min (x) and max (x) are the minimum and maximum values, respectively, in the dataset, Is normalized data.
4. The TFT-based crusher lining plate use state prediction method according to claim 1, wherein the specific steps of S21 are as follows: S211, extracting characteristics of time sequence data; Processing the time sequence data in a time sequence processing unit, firstly passing through two continuous one-dimensional convolution layers, then passing through a time attention layer, finally passing through a global maximum pooling layer to obtain an output vector with a fixed size of k, and finally obtaining a time sequence characteristic; s212, extracting features of scalar data; processing scalar data in a scalar processing unit, and expanding the dimension of each scalar into an array with a fixed size of k through a full connection layer, so that the scalar data and time sequence data are aligned in dimension, and thus scalar features are obtained; S213, combining the time sequence feature obtained in the S211 and the scalar feature obtained in the S212 into a matrix in the k-dimensional direction to form a combined feature vector, and then enhancing the association relationship between time sequence data and scalar data through a multi-head attention layer to obtain a final feature vector of the feature extraction module.
5. The TFT-based crusher lining plate use state prediction method according to claim 1, wherein the specific steps of S22 are as follows: S221, inputting the final feature vector into an LSTM unit of a prediction module to capture long-term and short-term dependency in time sequence data; S222, outputting the learned general features by the LSTM unit, respectively inputting the general features into three task-specific full-connection layers of the multi-task prediction unit, and further processing and learning to obtain prediction results of three tasks of the residual service life RUL of the lining plate, the wearing state of the lining plate and the running state of the crusher.
6. The TFT-based crusher lining plate use state prediction method according to claim 1, wherein the specific steps of S3 are as follows: s31, constructing a loss function Constructing the residual service life RUL of the main task lining board into a regression task, and the loss function of the regression task The calculation formula is as follows: ; where B is the number of samples in a single iteration calculation, And The true value and the predicted value of the j-th sample; auxiliary task liner wear state and crusher operation state are constructed as classification tasks, loss functions thereof The calculation formula is as follows: ; where B is the number of samples in a single iteration calculation, C is the number of categories, And The j-th sample real label and the vector of the prediction probability are respectively; The three losses are added according to the set weight size as a total loss: + ; Wherein, the The weight of the remaining life loss function for the primary service liner, The weight of the loss function for the subtask-liner wear state, The weight of a loss function of the running state of the secondary crusher of the subtask is given; S32, iterative training Inputting training data sets of time sequence data and scalar data into the model constructed in the step S2, calculating loss between a prediction result and a real label, executing a back propagation algorithm, and updating model parameters according to the loss; S33, model test And (3) inputting a test data set of the time sequence data and the scalar data into the model after iterative training for testing, if the test result is qualified, finishing the model training, otherwise, entering steps S32 and S33 to continue training and testing until the test result is qualified, and obtaining a trained model.
7. The TFT-based breaker bar usage state prediction method of claim 1, wherein S3 further comprises evaluating accuracy of model predictions using root mean square error and scoring functions; The loss function is used in a training stage, and the scoring function is used in a testing stage; The root mean square error is calculated as follows: ; where N is the total number of evaluation processes, And The true value and the predicted value of the ith sample respectively; the calculation formula of the scoring function is as follows: ; Wherein, the For the weight of the i-th prediction error, N is the total number of evaluation processes, Is a parameter for adjusting the score.
8. A TFT-based crusher liner service life prediction system for implementing the TFT-based crusher liner service state prediction method of claim 1, the system comprising: the data input module is used for acquiring data related to lining plate consumption and preprocessing the data to acquire time sequence data and scalar data; the feature extraction module comprises a time sequence processing unit, a scalar processing unit and a multi-head attention unit, wherein the time sequence processing unit is used for processing the time sequence data to obtain a time sequence feature vector, the scalar processing unit is used for processing the scalar data to obtain a scalar feature vector, and the multi-head attention unit is used for obtaining a total feature vector based on the time sequence feature vector and the scalar feature vector; and the prediction module is used for outputting a prediction result based on the total feature vector, wherein the prediction result comprises a lining plate abrasion state, a crusher running state and a lining plate residual service life.
9. The TFT-based crusher lining plate life prediction system of claim 8, wherein the preprocessing in the data input module comprises deletion of extraneous and anomalous data, data integration, and data normalization.
10. The TFT-based crusher liner life prediction system of claim 8, wherein the scalar data comprises liner shape, structure, material, hardness, crusher model, and crushing fraction, and wherein the time series data comprises ore hardness, grain size composition, crusher power, bin level, feed volume, feed grain size, discharge grain size, and liner replacement cycle.

Description

Method and system for predicting using state of crusher lining plate based on TFT Technical Field The invention belongs to the technical field of equipment life expectancy in equipment management, and particularly relates to a TFT (Temporal Fusion Transformer) -based crusher lining plate use state prediction method and system. Background The crusher is widely applied to material crushing equipment in industrial production, in the crushing process, the lining plate is used as a key part of the crusher, and is subjected to huge impact force and abrasion, and the lining plate is gradually abraded along with the increase of the service time, so that the ore processing efficiency and the equipment operation safety are directly affected. As the depth of exploitation of mineral resources increases and the ore grade decreases, the amount of non-minerals in the ore increases, leading to increasingly outstanding wear problems of the liner, and how to effectively predict the remaining service life RUL of the liner is critical to avoid unplanned outages and reduce maintenance costs. The service life of the liner is affected by a number of factors including ore conditions, equipment conditions, production conditions, etc. At present, the relevant literature on the method for predicting the service life of the lining plate of the crusher is as follows: In the prior art, the use state and the residual life of a crusher lining plate usually depend on-line Monitoring and prediction by installing sensors on the lining plate or equipment side, such as installing digital wireless sensors at the concave lining of a gyratory crusher and coupling with discrete meta-models to reconstruct thickness distribution and wear track (Ou et al, modelling of gyratory crusher LINER WEAR using A DIGITAL WIRELESS sensor, 2023), or an industrial scheme of embedding proprietary sensors into a lining plate casting to output wear degree in real time and guide replacement (WO 2020210875A1, WEAR SENSING LINER), and an on-line Monitoring method of inserting constant-speed wear probes through a shell and jointly evaluating the wear and impact position of the lining plate by utilizing ultrasonic flight time or vibration amplitude (WO 2022000072A1, monitor LINER WEAR IN industrial mills). The method is characterized in that sensors are arranged on the lining plate or equipment side, power supply and sealing problems, high calibration and maintenance cost, high dependence on the accuracy and the integrity of historical data, poor model generalization capability, difficult model-crossing migration and the like are faced under high-impact, high-abrasion and dust heat environments, and a shutdown measurement-based scheme cannot be continuously and online predicted, so that service life evolution under complex working conditions is difficult to reflect in time. In summary, in the prior art, there is no effective scheme for effectively predicting the service life of a lining plate according to the complex factors of the service life of the lining plate without using a lining plate sensor. Disclosure of Invention The invention aims to solve the problems of the prior art for predicting the service life of a lining plate of a crusher, and provides a TFT (Temporal Fusion Transformer) -based method and a TFT (Temporal Fusion Transformer) -based system for predicting the service state of the lining plate of the crusher. The TFT model is a time sequence prediction model combining a gate residual network and a multi-head attention mechanism, can process static and dynamic characteristics at the same time, and automatically captures the dependency relationship between a key time slice and an important variable. In the present invention, TFTs are used as the core backbone structure of the model, and an improved module for explicit alignment and multitasking prediction is added on this basis. The method comprises the steps of obtaining data related to lining plate consumption, preprocessing the data to obtain time sequence data and scalar data, reconstructing a life prediction model, and carrying out feature extraction and prediction on the obtained data to obtain a prediction result. According to the method, a lining plate sensor is not needed, so that the cost of a method for predicting the service life of the lining plate is reduced, the prediction accuracy is improved, the state and the service condition of the lining plate of the crusher can be analyzed in real time, the abnormal shutdown condition of equipment caused by the abrasion of the lining plate is greatly reduced, and the production and stable production are improved. The technical scheme of the invention is that the method for predicting the using state of the crusher lining plate based on the TFT comprises the following steps: S1, acquiring production data and lining plate technical data related to the consumption of a lining plate of a crusher, wherein the production data and the lining plate technical