Search

CN-121980380-A - Coiled material processing abnormality detection system and method based on multi-source sensing

CN121980380ACN 121980380 ACN121980380 ACN 121980380ACN-121980380-A

Abstract

The application provides a coil processing anomaly detection system and method based on multi-source sensing, which are used for carrying out weighted fusion on time domain statistical characteristics of a transmission mechanism node, texture characteristics of a surface travelling area, temperature gradient characteristics of a coil roller set and hot spot distribution characteristics on a coil processing production line to obtain fusion characteristic vectors in the coil processing, carrying out joint probability density estimation on normal working conditions of a coil processing Cheng Zhongkua sensing mode based on the fusion characteristic vectors to obtain normal working condition probability distribution, further calculating reconstruction errors and multivariable control diagram statistics between online multi-source sensing data and the normal working condition probability distribution on the coil processing production line, and locating anomaly sources of the coil processing line according to characteristic contribution degrees of all heterogeneous sensors when the reconstruction errors and the multivariable control diagram statistics continuously appointed times exceed preset anomaly threshold values. Based on the scheme, multi-source feature fusion and joint probability modeling can be realized.

Inventors

  • LIAN QUN
  • TIAN QING

Assignees

  • 深圳科宏健科技有限公司

Dates

Publication Date
20260505
Application Date
20260109

Claims (10)

  1. 1. A coil processing abnormality detection method based on multi-source sensing is characterized by comprising the following steps: synchronously arranging a plurality of heterogeneous sensors on a coil processing production line, and collecting vibration signals of a transmission mechanism node, surface image information of a surface travelling area and heat radiation distribution information of a coil roller set; Carrying out weighted fusion on time domain statistical characteristics of transmission mechanism nodes in the vibration signals, texture characteristics of surface travelling areas in the heat radiation data, temperature gradient characteristics of a coiled material roller group in the heat radiation distribution information and heat spot distribution characteristics to obtain fusion characteristic vectors in the coiled material manufacturing process; Carrying out joint probability density estimation on normal working conditions of a coil system Cheng Zhongkua sensing mode based on the fusion feature vector to obtain normal working condition probability distribution, and further calculating reconstruction errors and multivariate control diagram statistics between online multisource sensing data on a coil production line and the normal working condition probability distribution; and when the reconstruction errors and the multivariate control chart statistics continuously appointed times exceed a preset abnormal threshold, positioning an abnormal source of the coiled material production line according to the characteristic contribution degree of each heterogeneous sensor.
  2. 2. The method of claim 1, wherein weighting and fusing the time domain statistical features of the driving mechanism nodes in the vibration signal, the texture features of the surface traveling area in the heat radiation data, the temperature gradient features of the coil roller group in the heat radiation distribution information and the hot spot distribution features to obtain fused feature vectors in the coil process specifically comprises: performing noise reduction transformation on the vibration signal, and further extracting time domain statistical characteristics of nodes of the transmission mechanism; Performing texture enhancement on the surface image, and further extracting texture features of a local binary pattern; Respectively carrying out temperature calibration and space registration on the heat radiation data, and further extracting the temperature gradient characteristics and the hot spot distribution characteristics of the coiled material roller group; And carrying out weighted fusion on the time domain statistical features, the texture features, the temperature gradient features and the hot spot distribution features to obtain fusion feature vectors.
  3. 3. The method of claim 1, wherein performing joint probability density estimation on normal conditions of a coil system Cheng Zhongkua sensing mode based on the fusion feature vector, the obtaining a normal condition probability distribution specifically comprises: Pre-training the fusion feature vector to construct a depth representation space which is robust to working condition fluctuation and sensitive to abnormality; Mapping normal samples in the depth representation space to a structured latent space that is easy to estimate in density using a normalized stream model as a probability encoder; And fitting the joint probability distribution of the normal working condition samples in the structured potential space by adopting a nuclear density estimation method to obtain the normal working condition probability distribution.
  4. 4. The method of claim 1, wherein calculating reconstruction errors and multivariate control graph statistics between on-line multisource sensor data and the normal operating probability distribution on a web production line specifically comprises: The online acquired multisource sensing data is subjected to feature extraction and fusion and then is input into a trained variational self-encoder network; Calculating root mean square error between decoder output of the variation self-encoder network and the normal working condition probability distribution to obtain reconstruction error; And calculating the characterization value of the fusion feature vector of the multi-source sensing data in the potential feature space, and obtaining the statistic of the multivariable control chart relative to the negative log likelihood value of the probability distribution of the normal working condition.
  5. 5. The method of claim 1, wherein locating sources of anomalies in the web production line based on the characteristic contribution of each heterogeneous sensor comprises: Analyzing gradient contribution values of the reconstruction errors to each feature dimension in the original multi-source features; Aggregating gradient contributions of each characteristic dimension according to the attribution relation between the characteristics and the sensor to respectively obtain characteristic contribution degrees of the vibration, vision and thermal imaging sensor; and judging the sensor type with the highest characteristic contribution degree as an abnormal source, and further combining the physical meaning of the corresponding sensor characteristic to output an abnormal positioning result.
  6. 6. The method of claim 1, wherein the heterogeneous sensor comprises a non-contact laser vibrometer, a high resolution line camera, and a thermal infrared imager.
  7. 7. The method of claim 1, wherein the anomaly threshold is a critical value for determining whether the reconstruction error and the multivariate control graph statistic are in an anomaly state.
  8. 8. The coil processing abnormality detection system based on the multi-source sensing comprises an abnormality detection unit, and is characterized in that the abnormality detection unit comprises: the acquisition module is used for synchronously arranging a plurality of heterogeneous sensors on a coil processing production line and acquiring vibration signals of a transmission mechanism node, surface image information of a surface travelling area and heat radiation distribution information of a coil roller set; The processing module is used for carrying out weighted fusion on time domain statistical characteristics of the transmission mechanism nodes in the vibration signals, texture characteristics of surface advancing areas in the heat radiation data, temperature gradient characteristics of the coiled material roller group in the heat radiation distribution information and hot spot distribution characteristics to obtain fusion characteristic vectors in the coiled material manufacturing process; The processing module is also used for carrying out joint probability density estimation on the normal working condition of the coil system Cheng Zhongkua sensing mode based on the fusion feature vector to obtain normal working condition probability distribution, and further calculating reconstruction errors and multivariate control diagram statistics between online multisource sensing data on a coil production line and the normal working condition probability distribution; And the execution module is used for positioning an abnormal source of the coiled material production line according to the characteristic contribution degree of each heterogeneous sensor when the reconstruction error and the multivariable control diagram statistic continuously appointed times exceed a preset abnormal threshold value.
  9. 9. A computer device comprising a memory for storing a computer program and a processor for calling and running the computer program from the memory, such that the computer device performs the multi-source sensing-based coil process anomaly detection method of any one of claims 1 to 7.
  10. 10. A computer readable storage medium having instructions or code stored therein which, when executed on a computer, cause the computer to perform the multi-source sensing-based web manufacturing anomaly detection method of any one of claims 1 to 7.

Description

Coiled material processing abnormality detection system and method based on multi-source sensing Technical Field The application relates to the technical field of coiled material processing, in particular to a coiled material processing abnormality detection system and method based on multi-source sensing. Background The coil process is an industrial process for continuously producing sheet materials, such as metal strips, plastic films or paper, which is characterized in that raw materials are unreeled and processed for multiple times and finally wound into a coil by a winding device, and the process relies on precise coordination of a transmission system, tension control and temperature management, and minor anomalies of any link can cause product defects, production line shutdown and equipment damage. The traditional detection relies on a single type of sensor, so that the sensing dimension is narrow, abnormal characteristics of mechanical transmission state, surface microcosmic morphology, temperature field distribution and other multi-physical field coupling in the coil manufacturing process cannot be synchronously captured, complex faults caused by interweaving of various factors are difficult to deal with, and the model training is carried out by seriously relying on historical known abnormal samples, the detection logic is essentially that the matching and classification of the fault modes are carried out, the generalization recognition capability is lacking for novel fault modes or unknown abnormal variants which do not appear in the training set, the limitation of the single sensing source easily causes information blind areas, for example, vibration monitoring is difficult to find surface coating defects without obvious vibration response, the early temperature rise of an internal bearing cannot be directly judged, and the sensing mode of the one-sided is extremely easy to cause detection omission or misjudgment under complex working conditions. Therefore, how to realize multi-source feature fusion and joint probability modeling, so that the improvement of the accuracy and the positioning accuracy of anomaly detection becomes a difficult problem in the industry. Disclosure of Invention The application provides a coiled material processing abnormality detection system and method based on multi-source sensing, which can realize multi-source feature fusion and joint probability modeling, thereby improving the accuracy and positioning accuracy of abnormality detection. In a first aspect, the present application provides a method for detecting an abnormality in a web process based on multi-source sensing, including: synchronously arranging a plurality of heterogeneous sensors on a coil processing production line, and collecting vibration signals of a transmission mechanism node, surface image information of a surface travelling area and heat radiation distribution information of a coil roller set; Carrying out weighted fusion on time domain statistical characteristics of transmission mechanism nodes in the vibration signals, texture characteristics of surface travelling areas in the heat radiation data, temperature gradient characteristics of a coiled material roller group in the heat radiation distribution information and heat spot distribution characteristics to obtain fusion characteristic vectors in the coiled material manufacturing process; Carrying out joint probability density estimation on normal working conditions of a coil system Cheng Zhongkua sensing mode based on the fusion feature vector to obtain normal working condition probability distribution, and further calculating reconstruction errors and multivariate control diagram statistics between online multisource sensing data on a coil production line and the normal working condition probability distribution; and when the reconstruction errors and the multivariate control chart statistics continuously appointed times exceed a preset abnormal threshold, positioning an abnormal source of the coiled material production line according to the characteristic contribution degree of each heterogeneous sensor. In some embodiments, the weighting and fusing the time domain statistical feature of the transmission mechanism node in the vibration signal, the texture feature of the surface travelling area in the heat radiation data, the temperature gradient feature of the coiled material roller group in the heat radiation distribution information and the hot spot distribution feature to obtain a fused feature vector in the coiled material process specifically includes: performing noise reduction transformation on the vibration signal, and further extracting time domain statistical characteristics of nodes of the transmission mechanism; Performing texture enhancement on the surface image, and further extracting texture features of a local binary pattern; Respectively carrying out temperature calibration and space registration on the heat radiation data,