Search

CN-122020446-A - Paper machine fault diagnosis method based on multichannel grouping

CN122020446ACN 122020446 ACN122020446 ACN 122020446ACN-122020446-A

Abstract

The invention relates to the technical field of fault diagnosis of industrial equipment, and discloses a fault diagnosis method of a paper machine based on multichannel grouping, which comprises the steps of firstly preprocessing and resampling signals of different physical types, and grouping according to a vibration group, a pressure and current group and a temperature group; and inputting each group of time sequence windows into a multi-scale time-frequency dynamic attention module to extract time-frequency attention characteristics, obtaining time sequence representation with gentle change through constraint of a slow characteristic layer, reconstructing by a long-short-term memory self-encoder in the group, calculating a time sequence point reconstruction error and a window aggregation error, and finally combining a static threshold value and a self-adaptive dynamic threshold value based on normal window sliding statistics to finish fault judgment. The invention realizes high-precision modeling and self-adaptive anomaly detection of the multichannel complex timing signals, has the advantages of high diagnosis precision, low false alarm rate, small detection delay and the like, and is suitable for online monitoring and intelligent operation and maintenance of the paper machine.

Inventors

  • WU PING
  • WANG HUAIMIN
  • NI YUXUAN

Assignees

  • 浙江理工大学

Dates

Publication Date
20260512
Application Date
20251222

Claims (7)

  1. 1. A paper machine fault diagnosis method based on multichannel grouping is characterized by comprising the following steps of collecting vibration, pressure/current and temperature data of a paper machine in real time, grouping the data according to signal physical types, and processing each group of data in parallel according to the following steps: s1, data preprocessing, including filtering, resampling, deletion complement and normalization processing, is carried out, and a window to be inferred is constructed; s2, inputting a window to be inferred into an offline trained multi-scale time-frequency dynamic attention long-term and short-term memory self-encoder model, and firstly extracting attention perception characteristics through a multi-scale time-frequency dynamic attention module The slow feature layer then perceives the features to the attention Mapping to obtain slow characteristic sequence Then the reconstructed characteristic sequence is obtained through the intra-group long-short-term memory self-encoder Further, calculating a time sequence point reconstruction error and a window aggregation error for each time step t; S3, triggering an alarm if the number of the normal windows is smaller than the preset sliding window size W and the window aggregation error is larger than an initial dynamic threshold value, and triggering an alarm if the number of the normal windows is larger than or equal to the preset sliding window size W and the window aggregation error is larger than an adaptive dynamic threshold value.
  2. 2. A multi-channel packet based paper machine fault diagnosis method according to claim 1, characterized in that: The preprocessing comprises the steps of removing direct current components of vibration signals by using high-pass filtering, reserving effective frequency bands by using band-pass filtering, using low-pass filtering for pressure and current signals, using short-time moving average smoothing for temperature signals, then resampling, and forming a window to be inferred by the step length S=T/2 when a buffer area is accumulated to a complete window and the effective sample proportion is more than or equal to 95%.
  3. 3. A multi-channel packet based paper machine fault diagnosis method according to claim 2, characterized in that: the resampling comprises the steps of firstly carrying out anti-aliasing treatment by using a low-pass filter during downsampling, adopting linear interpolation for short-time deletion and interpolation for medium-time deletion during upsampling, and discarding a window with continuous deletion in a long time.
  4. 4. A method for diagnosing a paper machine failure based on multi-channel grouping according to any one of claims 1 to 3, characterized in that: the multi-scale time-frequency dynamic attention module comprises the steps of firstly extracting three features with different scales from an input sequence, adding the three features with different scales along a channel to obtain a fusion feature F, and then respectively calculating time domain attention mapping for the fusion feature F And frequency domain attention map Map the fusion feature F with time domain attention Obtaining weighted features by element-by-element multiplication Mapping the fusion feature F with the frequency domain attention Obtaining weighted features by element-by-element multiplication Features are then added And features Splicing in channel or feature dimension, fusing by convolution, outputting attention perception feature 。
  5. 5. A multi-channel packet based paper machine fault diagnosis method according to claim 4, wherein: The time sequence point reconstruction error is as follows: (7) the window aggregation error is as follows: (8) wherein T is a fixed window length.
  6. 6. A multi-channel packet based paper machine fault diagnosis method according to claim 5, wherein: the multi-scale time-frequency dynamic attention long-term and short-term memory self-encoder model is obtained through offline training by the following steps: Acquiring historical working condition data of a paper machine, including multichannel time sequence data of vibration, pressure/current and temperature data, dividing the data into a training set, a verification set and a test set according to time sequence, wherein the training set only comprises normal working condition samples, preprocessing each group of training set, verification set and test set respectively, inputting the training set into a multi-scale time-frequency dynamic attention long-short-term memory self-encoder model according to the group to obtain a reconstruction sequence, and calculating a time sequence point reconstruction error based on the training set and a window aggregation error based on the training set to obtain a static threshold based on the training set: wherein And The mean and standard deviation of the window aggregate error based on the training set, Verifying the performance of the model on a verification set after each round of training; And then inputting the test set into a multi-scale time-frequency dynamic attention long-short-term memory according to a group to obtain a reconstruction sequence from the encoder model, and calculating a time sequence point reconstruction error based on the test set and a window aggregation error based on the test set, thereby calculating the mean value mu and the standard deviation sigma of the window aggregation error of the normal sample in the test set.
  7. 7. A multi-channel packet based paper machine fault diagnosis method according to claim 6, wherein: The initial dynamic threshold is: (9) the adaptive dynamic threshold is: (10) Wherein, the And The mean value and standard deviation of window aggregation errors of the last W normal windows are respectively, The upper threshold is: 。

Description

Paper machine fault diagnosis method based on multichannel grouping Technical Field The invention belongs to the technical field of fault diagnosis of industrial equipment, and particularly relates to a fault diagnosis method of a paper machine based on multichannel grouping. Background The paper machine is used as the core equipment of the modern paper industry, the functions of the paper machine cover key process links such as pulp forming, squeezing and dewatering, drying and shaping, and the running stability of the paper machine directly influences the quality and the production efficiency of products. However, paper machines are complex in structure (comprising thousands of parts such as drying cylinders, press rolls, driving belts, vacuum pumps, etc.), and are prone to typical faults such as bearing wear, drying cylinder temperature unbalance, driving belt slipping, roller eccentricity, etc. under severe conditions of high temperature, high humidity and high mechanical load for a long time. Therefore, the real-time state monitoring and accurate fault diagnosis of the paper machine are realized, and the method has important significance for guaranteeing production safety and reducing operation and maintenance cost. In the current industrial scenario, existing methods for diagnosing faults of paper machines, such as principal component analysis (PRINCIPAL COMPONENT ANALYSIS, PCA) and Long Short-Term Memory (LSTM), have the following defects: 1. feature redundancy-multisource sensor data contains a large number of extraneous features, reducing model sensitivity. 2. The timing dependence is insufficient, the fault signal often appears as a long period gradual change, which is difficult to capture by conventional models. 3. The adaptability of the dynamic threshold is poor, and the fixed threshold is easy to cause false alarm and missing report. With the popularization of industrial internet of things (IIoT) technology, a fault diagnosis method based on machine learning is gradually applied to paper machine monitoring, but the following bottlenecks still exist: High dimensional data noise interference, namely paper machine multisource sensor data (vibration, temperature, pressure and the like) is high in dimension (usually more than 10 dimensions), is obviously influenced by environmental noise (such as workshop electromagnetic interference and mechanical vibration crosstalk), and is difficult to effectively distinguish key features from noise by a traditional dimension reduction method (such as PCA and t-distribution random neighborhood embedding (t-SNE)). The modeling of Long time sequence dependency is insufficient, namely a fault signal often shows a Long period gradual change characteristic, while the problem of gradient disappearance can be relieved by a traditional Long Short-Term Memory network model (LSTM), but the Memory capacity still can decline when facing a Long sequence, and the gradual change characteristic of the Long period is difficult to capture. The existing method mostly adopts a fixed threshold (such as 3 sigma principle) to judge abnormality, but the working condition of the paper machine dynamically changes along with the production task (such as the difference of the temperature set values of the drying cylinders corresponding to different paper types), and the fixed threshold is easy to cause false alarm or missing alarm. The fault positioning is fuzzy, that is, most diagnostic models can only judge that the fault occurs, and fault components (such as overheat of a drying cylinder and abnormality of an adjacent roller) cannot be accurately positioned, so that the maintenance efficiency is low. Disclosure of Invention The invention aims to solve the technical problem of providing a paper machine fault diagnosis method based on multichannel grouping, which is used for detecting multichannel time sequence data grouping of a paper machine through a multi-scale time-frequency dynamic attention long-short-term memory self-encoder (MS-MTFA-LSTM-AE) model and carrying out prediction alarm on faults by combining a dynamic threshold value. In order to solve the technical problems, the invention provides a paper machine fault diagnosis method based on multichannel grouping, which comprises the following steps of collecting vibration, pressure/current and temperature data of a paper machine in real time, grouping the data according to signal physical types, and processing each group of data in parallel according to the following steps: s1, data preprocessing, including filtering, resampling, deletion complement and normalization processing, is carried out, and a window to be inferred is constructed; s2, inputting a window to be inferred into an offline trained multi-scale time-frequency dynamic attention long-term and short-term memory self-encoder model, and firstly extracting attention perception characteristics through a multi-scale time-frequency dynamic attention module The slow featur