CN-122024731-A - Intelligent manufacturing workshop safety interaction control method and system based on large model

CN122024731ACN 122024731 ACN122024731 ACN 122024731ACN-122024731-A

Abstract

The invention relates to the field of industrial control, in particular to an intelligent manufacturing workshop safety interaction control method and system based on a large model, wherein the method comprises the steps of calculating the instantaneous energy of a sliding window and constructing the energy concentration degree by collecting sound signals of workshop environment, and judging local impact noise; the energy concentration is used as a priori factor to introduce self-adaptive Kalman filtering, the smooth energy change rate is tracked, the window length parameter is calculated, the dynamic window length is self-adaptively determined, after the short-time Fourier transform is used for noise reduction, the window length is multiplexed or reconstructed according to the frame similarity, and the pure voice is input into a voice recognition large model to realize safe interactive control. The invention solves the problems of easy noise tailing, insufficient frequency resolution and voice distortion of the traditional fixed window length, can effectively distinguish equipment impact noise from artificial voice instructions, improves signal processing precision and anti-interference capability, and is suitable for high-reliability voice safety interaction in complex industrial environments.

Inventors

ZHANG LIJIE
XU HAO
YUAN PENG
WANG JIANKANG
ZHU XINHUA
SHI PEITAO
LI BINGWEI

Assignees

山东青鸟工业互联网有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (8)

1. The intelligent manufacturing workshop safety interaction control method based on the large model is characterized by comprising the following steps of: Collecting sound signals of an intelligent manufacturing workshop, defining sliding windows for signal analysis and calculating the instantaneous energy of each sliding window; Constructing the energy concentration degree of the current frame sound signal based on the instantaneous energy of all sliding windows, and judging whether local impact noise exists in the current frame sound signal or not through the energy concentration degree to obtain a preliminary judging result; Based on the combination of the preliminary discrimination results, introducing the energy concentration degree as a priori feedback factor into the adaptive Kalman filtering, taking the difference value of the instantaneous energy of the adjacent window as the energy change rate of the sliding window, carrying out noise smoothing and impact inhibition tracking processing based on the energy change rate to obtain an estimated value of the energy change rate of the optimal window after interference elimination, and fusing the estimated value of the energy change rate and the energy concentration degree to obtain window length adjustment parameters of Fourier change so as to filter noise misjudgment caused by an operator loud instruction; calculating a dynamic window length of a current frame between a preset maximum window length and a preset minimum window length based on window length adjustment parameters, performing Fourier transform on sound signals by using the dynamic window length, denoising, and transmitting the denoised sound signals to a voice recognition large model at the rear end as a safe interaction input signal; And calculating a similarity evaluation value of the time domain characteristics of the current frame sound signal and the next frame sound signal, and based on the similarity evaluation value, judging that the next frame sound signal is subjected to multiplexing or reconstruction of the dynamic window length, so as to complete the safety interaction control of the intelligent manufacturing workshop.
2. The intelligent manufacturing shop safety interactive control method according to claim 1, wherein the step of calculating the instantaneous energy of each sliding window is: And sequentially summing the original sound signals in each sliding window, and dividing the summation result by the length of the sliding window to obtain the instantaneous energy of the sliding window.
3. The intelligent manufacturing shop safety interactive control method based on the large model according to claim 1, wherein the energy concentration degree is calculated by the following method: Calculating the ratio between the maximum value of the instantaneous energy of all sliding windows and the average value of the instantaneous energy of all sliding windows, carrying out exponential decay mapping on the standard deviation of the instantaneous energy of all sliding windows by using an exponential function, and taking the product of the ratio and the exponential decay mapping result as the energy concentration of the current frame sound signal.
4. The intelligent manufacturing shop safety interactive control method according to claim 1, wherein the step of judging whether the local impulsive noise exists in the current frame sound signal by the energy concentration degree is as follows: And judging that no local impact noise exists in the current frame sound signal, and obtaining a preliminary judging result for steady background noise or a steady continuous voice command signal of an operator according to the energy concentration value being larger than or equal to a preset concentration threshold value.
5. The intelligent manufacturing shop safety interactive control method based on the large model according to claim 1, wherein the window length adjustment parameters are obtained by the following steps: constructing a system state vector by using the difference value between the instantaneous energy of each window of the current frame and the instantaneous energy of the adjacent window, constructing a state transition matrix and an observation matrix based on the time sequence change characteristic of the energy, and taking the instantaneous energy of the sliding window as an observation value; Dynamically adjusting the observed noise covariance based on the energy concentration of the current frame, and adaptively adjusting the Kalman gain based on the observed noise covariance to complete tracking and smoothing of the energy change rate, thereby obtaining an optimal energy change rate estimated value after interference is removed; Taking the ratio between the optimal energy change rate estimation value and the average value of the instantaneous energy of all sliding windows of the current frame as the relative mutation rate, averaging the relative mutation rates of all windows, multiplying the average value with the energy concentration degree, and carrying out negative exponential mapping on the multiplication result to obtain window length adjustment parameters.
6. The intelligent manufacturing shop safe interaction control method based on the large model according to claim 1, wherein the calculation mode of the dynamic window length is as follows: The method comprises the steps of multiplying window length adjusting parameters with preset maximum window length to obtain a maximum window length weighted item, multiplying a difference value between 1 and the window length adjusting parameters with preset minimum window length to obtain a minimum window length weighted item, carrying out summation operation on the maximum window length weighted item and the minimum window length weighted item, carrying out downward rounding processing on a summation result, and taking the processed result as a dynamic window length corresponding to a current frame sound signal.
7. The intelligent manufacturing shop safety interactive control method according to claim 1, wherein the step of deciding to perform multiplexing or reconstruction of a dynamic window length on the next frame of sound signal based on the similarity evaluation value comprises: Calculating a similarity evaluation value of waveforms between a current frame sound signal and a next frame sound signal after denoising processing is completed, responding to the similarity evaluation value being larger than or equal to a preset similarity threshold value, judging that sound field states are highly similar, multiplexing the dynamic window length of the current frame as the window length in Fourier transform of the next frame to execute denoising, otherwise, judging that the sound field states are suddenly changed, and completing reconstruction of the dynamic window length according to the steps of instantaneous energy calculation, energy accuracy construction, window length adjustment parameter quantification and dynamic window length calculation on the next frame sound signal.
8. A large model based intelligent manufacturing shop safety interaction control system comprising a processor and a memory, the memory storing computer program instructions which when executed by the processor implement the large model based intelligent manufacturing shop safety interaction control method according to any of claims 1-7.

Description

Intelligent manufacturing workshop safety interaction control method and system based on large model Technical Field The present invention relates to the field of industrial control. In particular to an intelligent manufacturing workshop safety interaction control method and system based on a large model. Background The traditional mode is gradually replaced by a workshop production mode which is digitalized, networked and intelligentized into a core, and man-machine cooperation, equipment interconnection and intelligent scheduling become typical characteristics of an intelligent manufacturing workshop. The workshop safety interaction is used as a guarantee of running stability of the intelligent manufacturing workshop, key links are rapidly conveyed through production instructions, and the accuracy, the instantaneity and the anti-interference performance of the interaction are directly related to the production safety, the operation efficiency and the intelligent level of the workshop. However, on the site of the intelligent manufacturing workshop, various noise sources such as equipment operation, machining, air flow transmission and the like exist, voice interaction signals are easily interfered by background noise, so that the voice signals are distorted, the characteristics are fuzzy, production instruction transmission errors are caused, the workshop production efficiency is affected, misoperation of equipment and man-machine cooperation accidents are caused, and the workshop production safety is threatened. In the prior art, fourier transform is often used to process a voice signal to smooth noise, wherein short-time fourier transform is one of the most widely used signal processing methods, and the time-frequency domain analysis and noise filtering of the signal are realized by dividing the voice signal into a plurality of short-time frames and performing fourier transform respectively. However, the traditional short-time Fourier transform uses a fixed window length for signal framing processing, has technical contradiction difficult to reconcile, and becomes a core bottleneck of intelligent manufacturing workshop voice safety interaction, namely, although the long-time window can improve frequency resolution and accurately identify voice frequency characteristics, instantaneous noise energy can be diffused to adjacent voice frames due to a time domain smoothing effect to form a tailing phenomenon to cover voice formant characteristics and pollute effective voice signals, while the short-time window can accurately position instantaneous noise and reduce energy cross-frame diffusion, frequency resolution can be greatly reduced, voice fundamental frequency and background noise cannot be distinguished, voice core frequency characteristics are easy to lose, and voice distortion after denoising cannot meet the requirements of workshop voice safety interaction on signal definition and accuracy. Disclosure of Invention In order to solve the problems that the short-time Fourier transform with a fixed window length causes noise tailing and covers voice characteristics easily due to the time domain smoothing effect, the short-time window reduces frequency resolution and causes voice distortion after denoising, and the requirements of workshop voice safety interaction on signal definition and accuracy cannot be met, the invention provides the scheme in the following aspects. In a first aspect, a large model-based intelligent manufacturing shop safety interactive control method includes collecting sound signals of an intelligent manufacturing shop, defining sliding windows for signal analysis, and calculating instantaneous energy of each sliding window; the method comprises the steps of constructing the energy concentration of a current frame sound signal based on the instantaneous energy of all sliding windows, judging whether local impact noise exists in the current frame sound signal through the energy concentration to obtain a preliminary judging result, introducing the energy concentration as a priori feedback factor into a self-adaptive Kalman filter based on the preliminary judging result, taking the difference value of the instantaneous energy of adjacent windows as the energy change rate of the sliding windows, carrying out noise smoothing and impact suppression tracking processing based on the energy change rate to obtain an estimated value of the energy change rate of an optimal window after interference elimination, merging the estimated value of the energy change rate with the energy concentration to obtain a window length adjusting parameter of Fourier change, so as to filter noise misjudgment caused by an operator's loud command, calculating the dynamic window length of the current frame based on the window length adjusting parameter between a preset maximum window length and a minimum window length, carrying out Fourier transform on the sound signal and carrying out denoising processing, transmi