CN-121972764-A - TIG automatic welding pulse phase control method based on time sequence molten pool identification

CN121972764ACN 121972764 ACN121972764 ACN 121972764ACN-121972764-A

Abstract

The invention relates to a TIG automatic welding pulse phase control method based on time sequence molten pool identification. And acquiring a dynamic image of the molten pool through a high-speed visual sensor, and realizing multidimensional feature quantification of the curvature of the front edge of the molten pool, the gradient of a liquid-solid interface of the rear edge, the standard deviation of the central gray scale and the like by combining self-adaptive filtering and feature extraction. And outputting the normalized parent metal melt index, the molten pool expansion rate and the fillable confidence coefficient in real time by utilizing a lightweight CNN-LSTM hybrid network, and further generating a continuous molten pool dynamic phase angle based on an arctangent function model. After the phase angle and the current phase offset are mapped, microsecond-level pulse current phase adjustment is realized through double buffer processing of FPGA hardware, and arc stability is improved through S-shaped acceleration limiting measures. According to the scheme, the molten pool state sensing precision and the welding process control instantaneity are effectively improved, and the cooperative optimization of welding stability and filling adaptability is realized.

Inventors

GUO JUNFEI
LV ZHENWEN
DENG YUNYING
HUANG GUIPING

Assignees

广东博盈特焊技术股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260317

Claims (10)

1. A TIG automatic welding pulse phase control method based on time sequence molten pool identification specifically comprises the following steps: s1, acquiring a dynamic image sequence of a welding pool, and extracting pool profile characteristic data comprising pool front edge curvature, a pool rear edge liquid-solid interface gradient and a center area gray standard deviation based on the dynamic image sequence; S2, constructing a molten pool dynamic phase calculation unit based on time sequence correlation of gray standard deviation of a gradient concentric area of a liquid-solid interface of a trailing edge of the molten pool; S3, inputting the molten pool profile characteristic data into a molten pool dynamic phase resolving unit to obtain a base metal melt index, a molten pool expansion rate and a molten pool fillable confidence coefficient; s4, calculating a dynamic phase angle of the molten pool through an arctangent function based on the melt index of the base material, the expansion rate of the molten pool and the fillable confidence of the molten pool; s5, establishing a mapping relation between the dynamic phase angles of molten pools and the phase offset of the current; S6, based on the mapping relation, searching the phase offset of the pulse current; and S7, performing microsecond-level pulse current waveform phase shift control according to the pulse current phase shift amount.
2. The TIG automated welding pulse phase control method based on time-series puddle identification of claim 1, further comprising, after step S7: And S8, estimating a next frame phase angle by utilizing a first derivative fitted by a historical five-frame molten pool dynamic phase angle sequence, and starting the step transition of the S-shaped acceleration limiting curve when the difference value of adjacent phase offsets exceeds a threshold value aiming at the next frame phase angle.
3. The TIG automatic welding pulse phase control method based on the time sequence molten pool recognition according to claim 1, wherein the step S1 specifically comprises: Acquiring a dynamic image sequence of a welding pool, and extracting pool profile characteristic data comprising pool front edge curvature, a pool rear edge liquid-solid interface gradient and a center area gray standard deviation from the dynamic image sequence; Dynamically adjusting the size of a filtering window based on local variance analysis to generate a denoising molten pool image sequence; The method comprises the steps of carrying out Canny edge detection and cubic spline curve fitting on a molten pool front edge region in a denoising molten pool image sequence, calculating the curvature characteristic of the molten pool front edge, determining a curvature value based on a second derivative of a fitting curve, outputting a molten pool front edge curvature data sequence, and quantifying the geometric form change of the molten pool front edge; the method comprises the steps of (1) calculating directional gradient by using Sobel gradient operator on a molten pool trailing edge region in a denoised molten pool image sequence, extracting gradient characteristics of a trailing edge liquid-solid interface, analyzing a liquid-solid phase change boundary based on gradient amplitude and phase angle, and outputting a trailing edge liquid-solid interface gradient data sequence; And carrying out gray histogram statistical analysis on a molten pool central rectangular region in the denoised molten pool image sequence, calculating gray standard deviation characteristics of the central region, quantifying heat distribution uniformity based on pixel gray value standard deviation, and outputting a gray standard deviation data sequence of the central region.
4. The TIG automatic welding pulse phase control method based on time sequence molten pool recognition according to claim 1, wherein in step S2, the molten pool dynamic phase resolving unit is a lightweight CNN-LSTM hybrid network structure.
5. The TIG automatic welding pulse phase control method based on the time sequence molten pool recognition according to claim 1, wherein the step S2 specifically comprises: Modeling the time sequence correlation of the curvature of the front edge of the molten pool, the gradient of the liquid-solid interface of the rear edge and the gray standard deviation of the central area; inputting the molten pool profile characteristic data into the molten pool dynamic phase resolving unit to form a structured characteristic input vector; Based on the structured feature input vector, performing convolution feature extraction and long-term and short-term memory time sequence processing through a lightweight CNN-LSTM hybrid network, and calculating dynamic feature representation of a molten pool state; carrying out normalization calculation on the dynamic characteristic representation of the molten pool state to obtain a state characteristic parameter; and outputting a base material melt index, a molten pool expansion rate and a molten pool fillable confidence based on the state characteristic parameters.
6. The TIG automatic welding pulse phase control method based on the time sequence molten pool recognition according to claim 1, wherein the step S3 specifically comprises: normalizing the curvature of the front edge of the molten pool, the gradient of the liquid-solid interface of the rear edge and the gray standard deviation of the central area; Based on the standardized molten pool contour feature vector, performing spatial convolution operation through a convolution neural network layer of the lightweight CNN-LSTM hybrid network; inputting the molten pool morphological space feature map into a long-period memory network layer of a lightweight CNN-LSTM hybrid network, and performing time sequence dynamic modeling; performing full-connection layer mapping processing on the time sequence hidden state vector to generate an original parent metal melt index, an original molten pool expansion rate and an original molten pool fillable confidence coefficient; and performing normalization processing on the original parent metal melt index, the original molten pool expansion rate and the original molten pool fillable confidence.
7. The TIG automatic welding pulse phase control method based on the time sequence molten pool recognition according to claim 1, wherein the step S4 specifically comprises: based on the normalized parent metal melt index, the molten pool expansion rate and the molten pool fillable confidence level output by the molten pool dynamic phase calculation unit, three types of molten pool state characteristic parameters are obtained as input sources; performing product operation processing on the molten pool expansion rate and the molten pool fillable confidence coefficient to generate a molten pool expansion-filling combined characteristic quantity; performing angle calculation to process an original phase angle output value by applying an arctangent function based on the molten pool expansion-filling combined characteristic quantity and the base metal melting allowance characteristic quantity; Performing modulo 2 pi range mapping processing on the original phase angle, converting the original phase angle into a [0, 2 pi ] standard interval, and generating a normalized molten pool dynamic phase angle; and performing boundary continuity verification processing on the normalized molten pool dynamic phase angle, verifying phase angle output based on constraint conditions in the range of 0-2 pi radians, and generating a continuous time sequence state quantization signal representing the physical evolution progress of the molten pool.
8. The TIG automatic welding pulse phase control method based on the time sequence molten pool recognition according to claim 1, wherein the step S5 specifically comprises: Threshold partition processing is carried out on the characteristic data of the physical evolution stage of the molten pool, and a threshold interval of a dynamic phase angle of the molten pool is set based on the physical characteristics of the development stage of the molten pool; performing preset phase offset distribution processing based on the phase angle partition threshold set to generate a fixed offset mapping table; Carrying out sliding mode approach law calculation processing on a historical molten pool dynamic phase angle sequence, and calculating a phase offset dynamic correction value based on the product relation of a phase angle historical change rate and a preset approach rate parameter, wherein the phase angle historical change rate is obtained by carrying out first-order differential operation on the historical sequence, and the preset approach rate parameter is set according to the stability requirement of a welding process; and integrating the fixed offset mapping table with the sliding mode approach law dynamic correction unit, and selecting a corresponding mapping rule based on the real-time value of the dynamic phase angle of the molten pool so as to generate a complete phase angle-current phase offset mapping relation.
9. The method for controlling the pulse phase of the TIG automatic welding pulse based on the time sequence molten pool recognition according to claim 8, wherein the threshold interval is that a first threshold point corresponds to a zero radian of an ending point of an initial melting stage of a base material, a second threshold point corresponds to a starting point of a stable expansion stage of the molten pool and is one sixth radian of a circumference, a third threshold point corresponds to a starting point of a fillable stage of the molten pool and is one third radian of a circumference, and a fourth threshold point corresponds to a starting point of a transitional stage and is one half radian of a circumference.
10. The method for controlling pulse phase of TIG automatic welding based on time sequence molten pool recognition according to claim 1, wherein step S7 specifically comprises injecting the pulse current phase offset into a phase register of a PWM generator, and driving an IGBT driving circuit to implement pulse current waveform phase offset so as to match a pulse current high-energy phase or low-energy phase with a molten pool development phase.

Description

TIG automatic welding pulse phase control method based on time sequence molten pool identification Technical Field The invention relates to the technical field of welding control and intelligent manufacturing, in particular to a TIG automatic welding pulse phase control method based on time sequence molten pool identification. Background At present, in the field of pulse current control of TIG automatic welding, a plurality of mainstream technical schemes are available, especially aiming at the matching problem of pulse current phase and molten pool dynamic characteristics. In the prior art, most of the prior art adopts periodic synchronization, current period locking, fixed threshold judgment or control instruction issuing based on PLC/DSP, and part of the prior art is aided with visual detection or temperature feedback to perform rough state identification. In the aspect of molten pool evolution state identification, single-frame image feature judgment (such as molten pool area, gray level distribution center, edge width and the like) is commonly applied in the industry, and the result is used for setting pulse peak time, base value duration or duty ratio parameters, so that the self-adaptive adjustment of the welding process is realized to a certain extent. Part of advanced schemes also introduce multi-power source coupling, multi-channel feedback or take small hole size change, short circuit event and the like as current waveform trigger signals, so that the pulse current and the molten pool change realize relatively coarse correspondence. The prior art has the following outstanding problems that firstly, response delay issued by a traditional PLC or DSP stepping instruction is generally more than a plurality of milliseconds at a pulse current phase switching level, microsecond-level phase adjustment cannot be realized, so that current peak time and a molten pool physical stage are misplaced, secondly, complex dynamic rules in the time sequence development of the molten pool are ignored by relying on single or static characteristics as judgment basis, so that the mismatch of parent metal penetration and welding wire filling time influences welding seam compactness and forming quality, thirdly, under a multi-source feedback or period synchronization scene, high-precision closed-loop real-time adjustment capability is lacked, system self-adaptability is limited, the welding process is easily influenced by disturbance, quality consistency is difficult to ensure, and fourthly, the existing control methods of short circuit event triggering, duty cycle adjustment, double pulse misphasing method and the like are limited to discrete event driving, and continuous evolution of the molten pool and geometric physical corresponding relation with current waveforms are difficult to describe. Disclosure of Invention The application provides a timing molten pool identification-based TIG automatic welding pulse phase control method, which aims to solve one of the problems or one of the problems of the prior art mentioned in the background art. The application provides a TIG automatic welding pulse phase control method based on time sequence molten pool identification, which specifically comprises the following steps: s1, acquiring a dynamic image sequence of a welding pool, and extracting pool profile characteristic data comprising pool front edge curvature, a pool rear edge liquid-solid interface gradient and a center area gray standard deviation based on the dynamic image sequence; S2, constructing a molten pool dynamic phase calculation unit based on time sequence correlation of gray standard deviation of a gradient concentric area of a liquid-solid interface of a trailing edge of the molten pool; S3, inputting the molten pool profile characteristic data into a molten pool dynamic phase resolving unit to obtain a base metal melt index, a molten pool expansion rate and a molten pool fillable confidence coefficient; s4, calculating a dynamic phase angle of the molten pool through an arctangent function based on the melt index of the base material, the expansion rate of the molten pool and the fillable confidence of the molten pool; s5, establishing a mapping relation between the dynamic phase angles of molten pools and the phase offset of the current; S6, based on the mapping relation, searching the phase offset of the pulse current; and S7, performing microsecond-level pulse current waveform phase shift control according to the pulse current phase shift amount. The TIG automatic welding pulse phase control method based on the time sequence molten pool identification has the following beneficial effects: (1) By constructing a molten pool dynamic phase solution operator module, modeling a molten pool evolution process into a continuous physical phase system, extracting three key state indexes of a parent metal melt index, an expansion rate and a fillable confidence level from a molten pool profile sequence in real time b