CN-121978908-A - Intelligent control method and system for fixed torque electric gun based on tightening curve analysis

CN121978908ACN 121978908 ACN121978908 ACN 121978908ACN-121978908-A

Abstract

The invention discloses an intelligent control method and system of a fixed torque electric gun based on tightening curve analysis, which relate to the technical field of industrial automation control and comprise the steps of synchronously collecting mechanical signals and vibration signals in the tightening process, respectively extracting mechanical characteristics and vibration characteristics, fusing the mechanical characteristics and the vibration characteristics into a combined characteristic vector, and constructing a characteristic sequence matrix according to time sequence; the multi-scale sample sequence is obtained through multi-scale feature parallel processing, after information difference threshold screening, the optimized sample is input into a pre-trained multi-channel one-dimensional convolutional neural network to obtain probability distribution vectors, so that a traceable result is generated, confidence threshold evaluation is introduced based on the probability distribution, and diagnosis-based intelligent control of the diagnosis result is realized by combining different physical characteristic matching difference control strategies of faults. Through multi-source information fusion and intelligent decision, the early diagnosis and accurate control of faults in the screwing process are realized, and the reliability and the intelligent level of the assembly quality are improved.

Inventors

ZHANG MIAO
WEI WEI
LIU MINGZHONG
MU HAO
TIAN XIAONA
PAN BO
LI ZHEN
ZHANG HONGJUN

Assignees

贵州电网有限责任公司

Dates

Publication Date: 20260505
Application Date: 20251222

Claims (8)

1. The intelligent control method for the torque-fixed electric gun based on the tightening curve analysis is characterized by comprising the steps of acquiring a vibration signal under the same time stamp as the mechanical signal based on the mechanical signal obtained through the tightening curve analysis through a high-frequency sensor, and carrying out mechanical feature extraction and vibration feature extraction; splicing the mechanical features and the vibration features into joint feature vectors, constructing a feature sequence matrix according to time sequence, adopting a multi-scale feature parallel processing architecture to obtain N multi-scale sample sequences, screening the generated N samples through a preset information difference threshold to obtain N samples, inputting the N samples into a pre-trained multi-channel one-dimensional convolutional neural network model to obtain probability distribution vectors, and generating a tracing result; And introducing a confidence threshold according to the tracing result, performing differentiation treatment on physical characteristics of different faults, and matching a control strategy to realize triage control execution of different judging results.
2. The intelligent control method for the torque-setting electric gun based on the tightening curve analysis of claim 1, wherein the tightening curve refers to a two-dimensional relation curve of torque with angle change in the tightening process of a bolt in the prior art, and a time-synchronous torque sequence and an angle sequence are obtained from the curve.
3. The intelligent control method of the torque gun based on the tightening curve analysis of claim 2, wherein the mechanical feature extraction comprises the steps of preprocessing a mechanical signal by using trending, filtering, normalization and anomaly detection, systematically eliminating interference in the mechanical signal, and obtaining a high-quality signal; calculating the torque change rate, the angular speed and the instantaneous rigidity based on the high-quality signal and outputting the calculated torque change rate, the angular speed and the instantaneous rigidity in real time; and combining the torque change rate, the angular velocity and the instantaneous rigidity into an ordered array according to a fixed sequence at the same time point to obtain the mechanical feature vector.
4. The intelligent control method of the torque-fixed electric gun based on the tightening curve analysis of claim 3, wherein the vibration characteristic extraction comprises the steps of preprocessing an acquired vibration signal by using trending, filtering, normalization and anomaly detection, systematically eliminating interference of the vibration signal and obtaining a high-quality signal; Based on the high-quality signal, a one-dimensional time domain waveform signal is converted into a two-dimensional time-frequency spectrogram by utilizing short-time Fourier transform, so that time and frequency are distinguished, and feature extraction of a specific frequency band associated with a fault mode is carried out to obtain a vibration feature vector.
5. The intelligent control method of the torque gun based on the tightening curve analysis of claim 4, wherein the mechanical feature vector and the vibration feature vector are connected end to end in a feature dimension based on the same time stamp to generate the joint feature vector; Stacking according to the time sequence of the joint feature vectors to form a feature sequence matrix and performing dimension conversion, namely mapping the joint feature vector dimension into a channel dimension and mapping the time dimension into a sequence length dimension; Inputting the converted characteristic sequence matrix into a multi-scale time sequence characteristic analyzer, operating N different time sequence characteristic extraction operators through the analyzer, respectively analyzing an original sequence from different scales to generate N sample sequences, and forming an initial sample set; The N samples generated are screened, wherein the N samples are aligned according to the time dimension and the feature vector is remolded, the N samples are arranged into a sequence from thin to thick according to the dimension, and a new sequence is obtained through filtering; Calculating cosine similarity between each pair of adjacent samples in the new sequence, taking the cosine similarity as characteristic variation of each pair of adjacent samples, and screening out sample pairs with the characteristic variation larger than a threshold value through a preset information difference threshold value to obtain a final set containing n samples; Inputting the n screened samples into the multichannel one-dimensional convolutional neural network model; based on the multichannel one-dimensional convolutional neural network model, carrying out convolution and nonlinear activation on n parallel feature extraction channels to generate n groups of advanced feature maps, splicing along the feature channel dimension to form a global fusion feature tensor, carrying out information integration and nonlinear mapping on the global fusion feature tensor through a full connection layer, and generating a probability distribution vector through a Softmax output layer to serve as a tracing result.
6. The intelligent control method of the torque gun based on tightening curve analysis of claim 5, wherein the Softmax function ensures that the element value of the output vector is between 0 and 1 and the sum of all elements is 1 by performing exponential transformation and normalization on each element; the Softmax function is expressed as: Wherein i represents the current element index; An ith logic representing full link layer output; j represents traversing all element subscripts; represents the j-th logic of the full connection layer output and K represents the vector dimension.
7. The intelligent control method for a torque gun based on tightening curve analysis according to claim 6, wherein the confidence threshold comprises a high confidence threshold alpha and a low confidence threshold beta, and the highest probability value in the probability distribution vector is compared with the confidence threshold: When the highest probability value is more than or equal to alpha, judging that diagnosis is reliable, and executing an automatic control strategy; when the highest probability value is smaller than alpha and larger than beta, the diagnostic reliability is limited, and the detection and early warning are enhanced; When the highest probability value is smaller than beta, the diagnosis is judged to be unreliable, and manual intervention is needed.
8. An intelligent control system of a fixed torque electric gun based on tightening curve analysis and an intelligent control method of the fixed torque electric gun based on tightening curve analysis according to any one of claims 1-7 are characterized by comprising an acquisition unit, a control unit and a control unit, wherein the acquisition unit acquires vibration signals under the same time stamp with the mechanical signals based on the mechanical signals obtained by tightening curve analysis through a high-frequency sensor and performs mechanical feature extraction and vibration feature extraction; The tracing unit is used for splicing the mechanical features and the vibration features into joint feature vectors, constructing a feature sequence matrix according to time sequence, adopting a multi-scale feature parallel processing architecture to obtain N multi-scale sample sequences, screening the generated N samples through a preset information difference threshold to obtain N samples, and inputting the N samples into a pre-trained multi-channel one-dimensional convolutional neural network model to obtain probability distribution vectors so as to generate tracing results; And the control unit introduces a confidence coefficient threshold according to the tracing result, performs differentiation treatment on physical characteristics of different faults, and matches a control strategy to realize diagnosis-based control execution of different judging results.

Description

Intelligent control method and system for fixed torque electric gun based on tightening curve analysis Technical Field The invention relates to the technical field of industrial automation control, in particular to an intelligent control method and system for a constant torsion gun based on tightening curve analysis. Background In the field of industrial automation control, a torque control gun is used as a core tool for precise assembly, and an intelligent control technology of the torque control gun is becoming a key for improving the manufacturing efficiency and the product quality. Along with the promotion of intelligent manufacturing and industry 4.0, the demand of the assembly process for real-time fault diagnosis and self-adaptive control is rapidly increased, the traditional constant torsion electric gun tightening control method mainly depends on single torque curve monitoring, hidden association of multi-source signals is difficult to capture, fault types can not be identified and fault reasons can not be traced at an early stage only through judgment of a post-hoc threshold, problems of lag in fault identification, inaccurate tracing and the like are caused, and the requirements for early warning and accurate intervention in a high-precision assembly scene can not be met. In the field of torque-fixed electric gun control, the traditional methods are generally limited to threshold judgment of mechanical signals or simple statistical analysis, such as monitoring torque peak values or angle changes to trigger an alarm, and although the methods can realize basic qualification and disqualification judgment, dynamic information of auxiliary modes such as vibration signals and the like is ignored, diversified fault modes are difficult to distinguish, the traditional systems often have slow response when irreversible hard faults occur, and the optimization treatment of the treatable faults also depends on manual intervention, so that the problems of high misjudgment rate, low assembly efficiency, aggravation of potential safety hazards and the like exist. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides an intelligent control method for the fixed-torque electric gun based on tightening curve analysis, which solves the problem that the traditional fixed-torque electric gun cannot accurately identify and trace the fault type in the tightening process, and realizes early diagnosis and intelligent diagnosis control based on multi-mode fusion. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the invention provides an intelligent control method of a torque-fixed electric gun based on tightening curve analysis, which comprises the steps of acquiring a vibration signal under the same time stamp as the mechanical signal through a high-frequency sensor based on the mechanical signal obtained through the tightening curve analysis, and carrying out mechanical feature extraction and vibration feature extraction; splicing the mechanical features and the vibration features into joint feature vectors, constructing a feature sequence matrix according to time sequence, adopting a multi-scale feature parallel processing architecture to obtain N multi-scale sample sequences, screening the generated N samples through a preset information difference threshold to obtain N samples, inputting the N samples into a pre-trained multi-channel one-dimensional convolutional neural network model to obtain probability distribution vectors, and generating a tracing result; And introducing a confidence threshold according to the tracing result, performing differentiation treatment on physical characteristics of different faults, and matching a control strategy to realize triage control execution of different judging results. As an optimal scheme of the intelligent control method of the torque-fixing electric gun based on the tightening curve analysis, the tightening curve refers to a two-dimensional relation curve of torque changing along with angles in the tightening process of bolts in the prior art, and a time-synchronous torque sequence and an angle sequence are obtained from the curve. The mechanical characteristic extraction comprises preprocessing mechanical signals by using trending, filtering, normalization and anomaly detection, and systematically eliminating interference in the mechanical signals to obtain high-quality signals; calculating the torque change rate, the angular speed and the instantaneous rigidity based on the high-quality signal and outputting the calculated torque change rate, the angular speed and the instantaneous rigidity in real time; and combining the torque change rate, the angular velocity and the instantaneous rigidity into an ordered array according to a fixed sequence at the same time point to obtain the mechanical feature ve