Vehicle and method, circuit and system for monitoring a weld

Abstract

This application provides a method, circuit, and system for monitoring vehicles and welds. The method includes: during the welding process of a weld at a welding station, acquiring multiple electrical signals in the welding circuit where the weld is located; dividing the multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of the multiple electrical signals; extracting time-frequency features from target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld; and performing quality inspection on the weld based on the multiple time-frequency features to obtain inspection results, wherein the inspection results include the quality grade of the weld. This application solves the technical problem of low efficiency in weld quality inspection in related technologies.

Description

Technical Field

[0001] This application relates to the field of welding inspection, and more specifically, to a method, circuit, and system for monitoring vehicles and welds. Background Technology

[0002] With the continued growth in demand for lightweight new energy vehicles, aluminum alloy body welding technology has been widely used in vehicle manufacturing. During aluminum alloy welding, the quality of the weld directly determines the reliability and safety of the vehicle body structure.

[0003] Currently, welding quality inspection often employs methods such as post-weld visual inspection, X-ray flaw detection, and penetrant testing, which suffer from drawbacks such as complex inspection procedures and long inspection cycles.

[0004] There is currently no good solution to the above problems. Summary of the Invention

[0005] This application provides a method, circuit, and system for monitoring vehicles and welds, to at least solve the technical problem of low efficiency in weld quality inspection in related technologies.

[0006] According to one aspect of the embodiments of this application, a method for detecting weld seams is provided, comprising: during the welding process of a weld seam at a welding station, acquiring multiple electrical signals in the welding circuit where the weld seam is located; dividing the multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of the multiple electrical signals; extracting time-frequency features of target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld seam, wherein the signal values ​​of the multiple target signals in the target signal segments are within a preset range, and the duration for which the signal values ​​of the multiple target signals are within the preset range reaches a preset time; and performing quality detection on the weld seam based on the multiple time-frequency features to obtain a detection result, wherein the detection result includes the quality grade of the weld seam.

[0007] Optionally, time-frequency features are extracted from the target signal segments in the sub-signal segments at different time periods to obtain multiple time-frequency features of the weld, including: filtering multiple target signals in the target signal segments to obtain multiple transient features of multiple target signals; performing short-time Fourier transform on the multiple transient features of multiple target signals to obtain the time-frequency spectrum matrix of the target signal segments; and performing time-frequency statistical calculations based on the time-frequency spectrum matrix to obtain multiple time-frequency features.

[0008] Optionally, time-frequency statistical calculations are performed based on the time-spectrum matrix to obtain multiple time-frequency features, including: weighting the frequencies of each frame in the time-spectrum matrix to obtain frequency distribution features; calculating the mean of the time-spectrum entropy of each frame in the time-spectrum matrix to obtain time-spectrum entropy features; detecting time series of frequencies in the time-spectrum matrix that are lower than a first preset peak value, and obtaining short-circuit frequency features based on the time series; detecting at least one frequency in the time-spectrum matrix that is higher than a second preset peak value, and calculating the ratio of the energy of at least one frequency to the total energy corresponding to the time-spectrum matrix to obtain energy proportion features, wherein the first preset peak value is lower than the second preset peak value; calculating the mean square error between the spectra of adjacent frames in the time-spectrum matrix to obtain time-domain variance features; wherein the multiple time-frequency features include time-frequency distribution features, time-spectrum entropy features, short-circuit frequency features, energy proportion features, and time-domain variance features.

[0009] Optionally, the weld quality is inspected based on multiple time-frequency features to obtain inspection results, including: determining the detection threshold ranges corresponding to the multiple time-frequency features, wherein the time-frequency features and the detection threshold ranges correspond one-to-one; and obtaining the inspection results based on the detection threshold ranges corresponding to the multiple time-frequency features and the multiple time-frequency features.

[0010] Optionally, based on the detection threshold ranges corresponding to multiple time-frequency features and the multiple time-frequency features, the detection result is obtained, including: based on the detection threshold ranges corresponding to multiple time-frequency features, determining whether the multiple time-frequency features are abnormal, and obtaining the number of abnormal time-frequency features, wherein the abnormal time-frequency features are not within the detection threshold range corresponding to the abnormal time-frequency features; and obtaining the detection result based on the number of abnormal time-frequency features.

[0011] Optionally, based on the number of anomalies with abnormal time-frequency characteristics, the detection results are obtained, including: if the number of anomalies is less than or equal to a first number, the quality level of the weld is determined to be level one; if the number of anomalies is greater than the first number but less than a second number, the quality level of the weld is determined to be level two, wherein the first number is less than the second number and the quality of level one is less than that of level two; if the number of anomalies is greater than or equal to the second number, the quality level of the weld is determined to be level three, wherein the quality of level two is less than that of level three.

[0012] Optionally, the detection threshold ranges corresponding to multiple time-frequency features are determined, including: acquiring multiple historical time-frequency features of historical welds; for any one time-frequency feature, performing statistics on the historical time-frequency features corresponding to that one time-frequency feature to obtain the statistical value of that one time-frequency feature; and determining the detection threshold range of that one time-frequency feature based on the statistical value.

[0013] Optionally, the electrical signal includes a current signal; based on the signal values ​​of multiple electrical signals, the multiple electrical signals are divided into sub-signal segments of different time periods, including: a first acquisition time for detecting the current signal rising from its initial value to the arc ignition detection threshold; a second acquisition time for acquiring the current signal rising from the arc ignition detection threshold and stabilizing at a preset state, wherein the preset state is that the signal value of the current signal is within a preset range, and the duration of the signal value of the current signal being within the preset range reaches a preset time, and the second acquisition time is later than the first acquisition time; a third acquisition time for detecting the current signal value being lower than the arc extinction detection threshold when the electrical signal is stable at the preset state, wherein the third acquisition time is later than the second acquisition time; a fourth acquisition time for determining the current signal decreasing from the arc extinction detection threshold to its initial value, wherein the fourth acquisition time is later than the third acquisition time; based on the first acquisition time, the second acquisition time, the third acquisition time, and the fourth acquisition time, the multiple electrical signals are segmented to obtain sub-signal segments of different time periods.

[0014] Optionally, the above method further includes: counting the number of welds that have been detected; when the number of welds that have been detected reaches the update number, obtaining a preset number of target welds from the welds that have been detected based on the quality level; and obtaining a new detection threshold range based on the time-frequency characteristics of the target welds.

[0015] Optionally, the above method further includes: associating and storing the detection results, target signal, and multiple time-frequency features with the vehicle identification and weld code of the weld, respectively.

[0016] Optionally, the above method also includes: controlling the welding station based on the test results.

[0017] Optionally, based on the detection results, the welding station can be controlled, including: if the quality level is less than or equal to the second level, controlling the welding station to continue welding the next weld; if the quality level is greater than the second level, outputting an early warning message and controlling the welding station to stop welding.

[0018] According to another aspect of the embodiments of this application, a weld detection circuit is also provided, optionally including: a data acquisition circuit connected to the weld at the welding station, used to acquire multiple electrical signals in the welding circuit where the weld is located during the welding process of the weld at the welding station; a controller connected to the data acquisition circuit, used to divide the multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of the multiple electrical signals, extract time-frequency features of target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld, and perform quality detection on the weld based on the multiple time-frequency features to obtain a detection result, wherein the signal values ​​of multiple target signals in the target signal segments are within a preset range, and the duration for which the signal values ​​of multiple target signals are within the preset range reaches a preset time, and the detection result includes the quality level of the weld.

[0019] Optionally, the acquisition circuit includes: a sensor connected to the weld seam for acquiring multiple analog signals in the welding circuit; a filter located between the sensor and the data acquisition card for filtering the multiple analog signals to obtain multiple processed analog signals; and a data acquisition card located between the filter and the controller for converting the multiple processed analog signals into multiple electrical signals.

[0020] Optionally, the sensors include: a voltage sensor connected in parallel with the weld seam for acquiring multiple voltage signals; and a current sensor connected in series with the weld seam for acquiring multiple current signals; wherein the multiple analog signals include multiple voltage signals and multiple current signals.

[0021] According to another aspect of the embodiments of this application, a weld inspection system is also provided, comprising: a signal acquisition module, used to acquire multiple electrical signals in the welding circuit where the weld is located during the welding process of the weld at the welding station; a feature transformation module, used to divide the multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of the multiple electrical signals, and to extract time-frequency features of target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld, wherein the signal values ​​of the multiple target signals in the target signal segments are within a preset range, and the duration for which the signal values ​​of the multiple target signals are within the preset range reaches a preset time; and a quality judgment module, used to perform quality inspection on the weld based on the multiple time-frequency features to obtain an inspection result, wherein the inspection result includes the quality level of the weld.

[0022] According to another aspect of the embodiments of this application, a vehicle is also provided, including: a body, wherein welds on the body are detected by the detection method in various embodiments of this application, or by the detection circuit in various embodiments of this application, or by the detection system in various embodiments of this application.

[0023] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0024] According to another aspect of the embodiments of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0025] According to another aspect of the embodiments of this application, a computer program product is also provided, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0026] According to another aspect of the embodiments of this application, a computer program is also provided, which, when executed by a processor, implements the methods of the various embodiments of this application.

[0027] This application involves acquiring multiple electrical signals within the welding circuit of the weld seam during the welding process at a welding station. Based on the signal values ​​of these signals, the signals are divided into sub-signal segments of different time periods. Time-frequency features of target signal segments within these sub-signal segments are extracted to obtain multiple time-frequency features of the weld seam. Based on these multiple time-frequency features, the weld seam is inspected to obtain the inspection results. This application acquires multiple electrical signals during the welding process at a welding station, automatically identifies and extracts target signal segments based on their signal values, and extracts multi-dimensional time-frequency features from these segments to obtain multiple time-frequency features for weld seam quality inspection. By utilizing these multiple time-frequency features corresponding to the target signal segments for quality inspection, the application achieves high-efficiency inspection, thereby improving the technical efficiency of weld seam inspection and solving the technical problem of low efficiency in weld seam quality inspection in related technologies. Attached Figure Description

[0028] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0029] Figure 1 This is a flowchart of a weld inspection method according to an embodiment of this application;

[0030] Figure 2 This is a schematic diagram of a weld detection circuit according to an embodiment of this application;

[0031] Figure 3 This is a schematic diagram of a weld inspection system according to an embodiment of this application;

[0032] Figure 4 This is a schematic diagram of an optional weld inspection method according to an embodiment of this application;

[0033] Figure 5 This is a schematic diagram of an optional weld detection circuit according to an embodiment of this application;

[0034] Figure 6 This is a schematic diagram of an optional weld inspection system according to an embodiment of this application;

[0035] Figure 7 This is a schematic diagram of a weld inspection device according to an embodiment of this application. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0037] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0038] According to an embodiment of this application, a method embodiment for inspecting welds is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0039] This embodiment provides a method for inspecting welds. Figure 1This is a flowchart of a weld inspection method according to an embodiment of this application, such as... Figure 1 As shown, the process includes the following steps:

[0040] Step S102: During the welding process of the weld seam at the welding station, multiple electrical signals in the welding circuit where the weld seam is located are collected.

[0041] The aforementioned welding station refers to the physical area in a vehicle manufacturing production line used for welding aluminum alloy car bodies. The welding station is equipped with a welding robot, welding power source, wire feeding mechanism, shielding gas supply system, and integrated detection devices. Within this specific space, the welding robot precisely positions and moves the car body components according to a preset trajectory, while the welding power source provides stable arc energy, converting electrical energy into heat energy through the welding torch to melt the base material and filler material to form a molten pool.

[0042] The welding station integrates a high-precision sensor network to capture electrical signals such as current and voltage in real time during the welding process. These signals directly reflect the stability of the arc combustion, the dynamic behavior of the molten pool, and the physical process of short-circuit transition. An integrated online inspection system enables real-time, non-destructive online inspection at the welding station. The layout design of the welding station fully considers the electromagnetic compatibility of signal acquisition, the sensor installation positions to avoid spatter damage, and the real-time response capability of the vehicle's control system. This ensures that high-quality welding and high-precision quality inspection can be completed simultaneously under high-speed automated production, thereby guaranteeing the integrity and safety of the lightweight body structure of new energy vehicles.

[0043] The aforementioned weld refers to the connection area formed at the joint of aluminum alloy body parts using gas metal arc welding (GMAW). The weld is a critical load-bearing component for the structural strength of the vehicle body, and its quality determines the vehicle's safety, sealing, and corrosion resistance. During welding, as the welding torch moves, the molten aluminum alloy cools and solidifies under the protection of the shielding gas, forming a metallic connection with a specific geometry and microstructure—the weld. The quality of the weld depends not only on its smoothness and width but, more importantly, on the presence of internal defects such as porosity, cracks, lack of fusion, or slag inclusions. These defects are often difficult to detect through routine visual inspection but can propagate due to fatigue loads during vehicle operation, leading to structural failure.

[0044] The weld is the object of data acquisition and analysis in this embodiment. The physical state of the weld is indirectly mapped through the electrical signals in the welding circuit. Therefore, the detection of the weld is essentially the analysis and inversion of the electrical signal characteristics of the welding thermal and metallurgical processes. By performing real-time quality rating on each weld through an online detection system, full life-cycle quality control of the vehicle body manufacturing process can be achieved, ensuring that every connection meets vehicle quality requirements.

[0045] The aforementioned welding circuit refers to the electrical system that connects the welding power source, welding torch, workpiece, and grounding clamp, forming a complete current flow path. The welding circuit is the energy transfer channel through which electrical energy is converted into heat energy and ultimately forms the weld joint. In the welding circuit, the welding power source outputs specific current and voltage waveforms. The current is conducted through the welding wire to the electric arc, then from the arc to the workpiece, and finally returns to the power source through the ground wire, forming a closed loop.

[0046] The electrical signals in the welding circuit are a direct representation of the physical processes of the electric arc, and they fluctuate dynamically with changes in welding parameters such as arc length, molten pool state, and short-circuit transfer frequency. Analyzing the time-frequency characteristics of these electrical signals can reveal the stability of arc combustion, the mode of droplet transfer, and the solidification behavior of the molten pool, providing a non-contact basis for judging weld quality.

[0047] The aforementioned electrical signals refer to signals reflecting changes in the physical state of the welding process within the welding circuit. These signals include, but are not limited to, current and voltage signals. Electrical signals are the core data source for online welding quality monitoring. Under high-frequency sampling, electrical signals exhibit complex time-varying characteristics, including steady-state components, abrupt changes caused by short-circuit transitions, and noise interference. By preprocessing the electrical signals to retain transient characteristics, and then performing time-frequency transformation, deeper information hidden within the signals can be extracted. The acquisition accuracy, sampling rate, and synchronization of the electrical signals directly affect the accuracy of subsequent feature extraction, thus determining the reliability of quality assessment. For example, in aluminum alloy welding, the electrical signals contain fault mode information; for instance, spatter can cause an abnormally high proportion of high-frequency energy, and arc instability can lead to an increase in spectral entropy.

[0048] In one optional embodiment, current and voltage sensors are used to simultaneously acquire multiple electrical signals in the welding circuit. A current sensor is integrated in series at the welding power supply output in the welding circuit, enabling the capture of transient current changes at the millisecond or even microsecond level. Simultaneously, a voltage sensor is connected in parallel between the welding torch and the workpiece to monitor voltage fluctuations across the arc in real time. Acquiring multiple electrical signals through these sensors provides a high-fidelity raw data foundation for subsequent time-frequency feature extraction.

[0049] In another alternative embodiment, a Rogowski coil is placed around the welding torch cable or workpiece cable, and the current value is inferred by measuring the induced electromotive force generated by the rate of change of current. To compensate for integration errors, a high-precision active integrator circuit is connected to the rear end of the Rogowski coil to restore the differential signal to the true current waveform. The voltage signal is attenuated by a high-voltage divider resistor network. The divider resistors are composed of low-temperature drift precision resistors to ensure the stability of the voltage division ratio. The attenuated low-voltage signal is processed by an isolation amplifier to match the input range of the data acquisition card.

[0050] In another alternative embodiment, fiber optic current sensors are placed near the nodes of the welding loop. Utilizing the Faraday gyro effect, when polarized light passes through an optical fiber placed in the magnetic field generated by the welding current, its polarization plane rotates. The rotation angle is proportional to the current magnitude. By detecting the rotation angle of the polarization plane, the current signal can be accurately obtained. Additionally, using the principles of capacitive coupling or electric field induction, fiber optic voltage sensors measure changes in the electric field and then calculate the voltage signal. The optical signal is transmitted through the optical fiber to a demodulator, which converts the optical signal into an electrical signal and performs digital processing. The fiber optic current and voltage sensors are small and lightweight, making them easy to embed into the confined space of the welding robot's end effector without increasing the mechanical load.

[0051] Step S104: Based on the signal values ​​of multiple electrical signals, divide the multiple electrical signals into sub-signal segments of different time periods.

[0052] In one optional embodiment, a fixed time window length is preset, such as 0.5 seconds or 1 second. The continuous electrical signal is segmented into equal-length sub-signal segments by moving the sliding window frame-by-frame across the acquired continuous time series. Within each time window, the average, effective, or maximum values ​​of the current and voltage signals are calculated, and this time window is defined as a sub-signal segment with specific physical meaning. Alternatively, the window length can be dynamically adjusted based on process parameters, or secondary clustering analysis can be performed on the signals within each window in subsequent steps to identify sub-signal segments that truly represent a specific welding stage (such as arc stabilization).

[0053] In another alternative embodiment, the transition points of the welding stages are identified by calculating the first derivative (rate of change) of the current signal. The start and end times of each stage are dynamically determined by monitoring the amplitude and sign changes of the slope, combined with integration criteria (such as the duration of a sustained positive or negative slope). For example, in pulsed MIG welding, the switching between the base current and the pulse current also manifests as a change in the slope; this method can automatically identify these periodic changes, thus allowing for more precise segmentation of the electrical signal. Furthermore, setting a hysteresis interval for the slope change can prevent oscillating segmentation near the threshold caused by noise.

[0054] Step S106: Extract time-frequency features from the target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld.

[0055] Among them, the signal values ​​of multiple target signals in the target signal segment are within a preset range, and the duration for which the signal values ​​of multiple target signals are within the preset range reaches a preset time.

[0056] The aforementioned target signal segment refers to a subset of signals automatically segmented from a complete sub-signal segment based on current amplitude characteristics, representing a stable arc-burning phase. The target signal segment is the fundamental unit for time-frequency feature extraction. During welding, the arc-ignition phase involves a transient surge in current, and the arc-extinguishing phase involves current decay to zero. Neither of these phases constitutes a normal stable weld pool formation stage. Feature extraction in these phases would introduce significant non-steady-state noise, leading to feature distortion. Therefore, using target signal segments to reflect the stable combustion of the arc and the regular short-circuit transition of the molten droplets provides good periodicity and stability. Time-frequency analysis of the target signal segment can more accurately capture key features such as the arc's dominant frequency energy distribution and spectral entropy, avoiding interference from the transient processes of arc ignition and extinguishing, thereby improving the sensitivity and accuracy of weld defect detection.

[0057] The aforementioned time-frequency characteristics refer to multidimensional statistics extracted from the combined time and frequency domain perspectives of a signal through time-frequency analysis of target signal segments, which characterize the physical state of the welding process. Time-frequency characteristics include, but are not limited to, time-frequency energy centroid frequency, time-spectral entropy, short-circuit transition frequency, high-frequency energy proportion, and time-domain variance of the time spectrum. These characteristics describe the concentration of energy distribution in the frequency domain, the uniformity of frequency components, the periodicity of short-circuit events, and dynamic fluctuation characteristics of the target signal. Multiple time-frequency characteristics offer higher information density and stronger defect characterization capabilities, enabling the capture of short-term transient anomalies and thus improving the sensitivity of welding defect detection.

[0058] In one optional embodiment, a deep learning model is employed to achieve end-to-end automatic extraction of time-frequency features. A convolutional neural network or recurrent neural network model is constructed, taking the target signal segment as input. The deep learning model automatically learns deep time-frequency features in the signal through multiple convolutional and pooling operations, thereby capturing complex patterns and implicit correlations in the signal that are difficult to identify by the naked eye. The final fully connected layer outputs multiple time-frequency features. During the training process of the deep learning model, a large amount of labeled qualified and unqualified weld data is used for supervised learning, enabling the deep learning model to learn to distinguish the time-frequency features corresponding to different quality levels.

[0059] In another optional embodiment, an adaptive time-frequency analysis method combining empirical mode decomposition (EMD) and Hilbert-Huang transform (HWT) is employed to obtain multiple time-frequency features of the weld. EMD is performed on the target signal segment, decomposing the nonlinear and non-stationary signal into several intrinsic mode functions (EMFs) and residual components. Each EMF represents the inherent oscillation mode of the signal at different time scales. Subsequently, a Hilbert transform is performed on each EMF to obtain its instantaneous amplitude and instantaneous frequency, constructing a Hilbert marginal spectrum. Multiple time-frequency features are extracted by statistically analyzing the energy distribution, instantaneous frequency mean, and variance of each EMF.

[0060] Step S108: Based on multiple time-frequency features, perform quality inspection on the weld and obtain the inspection results.

[0061] The test results include the quality grade of the weld.

[0062] The aforementioned detection results refer to quantitative or qualitative conclusions about the weld quality status based on multiple extracted time-frequency features. Detection results may include, but are not limited to, quality grades and corresponding risk indicators. The generation of detection results relies on the comprehensive analysis of multi-dimensional time-frequency features to achieve graded early warning of welding defects, thereby distinguishing between minor fluctuations and serious faults. This grading mechanism provides more nuanced quality information. The real-time output of detection results and the linkage with the control system achieve a closed-loop feedback from detection to control, ensuring that non-conforming products do not flow into the next process.

[0063] In one optional embodiment, a classifier based on a support vector machine (SVM) is used to determine the detection result. A supervised learning algorithm is employed, using historical weld time-frequency feature data labeled with quality levels (qualified, warning, unqualified) to train the SVM model. The low-dimensional feature space is mapped to a high-dimensional feature space, and a hyperplane is found in the high-dimensional space to separate samples of different quality levels as much as possible. During the detection phase, multiple time-frequency features of the weld are input into the trained SVM, which outputs the quality level of the weld and the decision function value, thus obtaining the detection result.

[0064] In another alternative embodiment, the detection results are obtained based on fuzzy logic inference. For example, a fuzzy logic system is constructed, transforming the numerical values ​​of each time-frequency feature into fuzzy linguistic variables (such as "normal," "slightly abnormal," and "severely abnormal") through a membership function. Next, a fuzzy rule base is defined, with rules derived from expert experience or historical data analysis. Finally, the conclusions of all rules are aggregated using a fuzzy inference engine, and a defuzzification method (such as the centroid method) is used to obtain a clear quality level output. By adjusting the shape of the membership function and the fuzzy rules, different process requirements can be flexibly adapted.

[0065] In another alternative embodiment, unsupervised anomaly detection based on isolated forests determines the detection results. Multiple isolated trees are constructed by randomly selecting features and split points. The average path length of each sample is calculated; a shorter path indicates a higher likelihood of an anomaly. A threshold is set based on the path length to classify samples into normal and abnormal categories, or further clustering can be used to distinguish anomalies of different severity. By calculating the contribution of each time-frequency feature to the isolated path length, the specific features causing the anomaly can also be identified, aiding in fault diagnosis and quickly screening for obvious abnormal welds.

[0066] This application involves acquiring multiple electrical signals within the welding circuit of the weld seam during the welding process at a welding station. Based on the signal values ​​of these signals, the signals are divided into sub-signal segments of different time periods. Time-frequency features of target signal segments within these sub-signal segments are extracted to obtain multiple time-frequency features of the weld seam. Based on these multiple time-frequency features, the weld seam is inspected to obtain the inspection results. This application acquires multiple electrical signals during the welding process at a welding station, automatically identifies and extracts target signal segments based on their signal values, and extracts multi-dimensional time-frequency features from these segments to obtain multiple time-frequency features for weld seam quality inspection. By utilizing these multiple time-frequency features corresponding to the target signal segments for quality inspection, the application achieves high-efficiency inspection, thereby improving the technical efficiency of weld seam inspection and solving the technical problem of low efficiency in weld seam quality inspection in related technologies.

[0067] Optionally, time-frequency features are extracted from the target signal segments in the sub-signal segments at different time periods to obtain multiple time-frequency features of the weld, including: filtering multiple target signals in the target signal segments to obtain multiple transient features of multiple target signals; performing short-time Fourier transform on the multiple transient features of multiple target signals to obtain the time-frequency spectrum matrix of the target signal segments; and performing time-frequency statistical calculations based on the time-frequency spectrum matrix to obtain multiple time-frequency features.

[0068] In one alternative embodiment, an adaptive median filter or a similar non-smooth filtering algorithm is used to remove random pulse noise and background interference in the welding circuit, while retaining the transient change details in the electrical signal that reflect physical processes such as short-circuit transition and molten droplet splashing, i.e., transient features. This achieves a balance between noise suppression and transient feature retention, ensuring that key physical events of the target signal (such as the current drop at the moment of short circuit) are not obscured.

[0069] Then, short-time Fourier transforms are performed on multiple transient features of multiple target signals, such as through windowing (e.g., Hanning window) and fast Fourier transform, to convert them into a two-dimensional matrix representation that simultaneously contains time and frequency information, i.e., the time spectrum matrix. By selecting a window length covering a specific short-circuit transition period and an appropriate frame shift step size, a time spectrum matrix that can clearly show the trajectory of energy distribution changes over time is generated, thereby comprehensively characterizing arc stability, spatter intensity, and dynamic fluctuations of the molten pool.

[0070] Finally, time-frequency statistical calculations are performed based on the time-spectrum matrix, and time-frequency features with clear physical meaning are extracted using mathematical statistical methods. The time-spectrum data is reduced to a small number of key indicators to facilitate efficient quality assessment, while ensuring that these indicators comprehensively cover welding quality across different dimensions, such as arc frequency, distribution uniformity, short-circuit period stability, spatter intensity, and dynamic fluctuations. For example, by using weighted averaging, entropy calculation, peak detection, and variance statistics, the energy distribution patterns and trends in the time-spectrum matrix are quantified, transforming the abstract time-spectrum into calculable and comparable features.

[0071] Optionally, time-frequency statistical calculations are performed based on the time-spectrum matrix to obtain multiple time-frequency features, including: weighting the frequencies of each frame in the time-spectrum matrix to obtain frequency distribution features; calculating the mean of the time-spectrum entropy of each frame in the time-spectrum matrix to obtain time-spectrum entropy features; detecting time series of frequencies in the time-spectrum matrix that are lower than a first preset peak value, and obtaining short-circuit frequency features based on the time series; detecting at least one frequency in the time-spectrum matrix that is higher than a second preset peak value, and calculating the ratio of the energy of at least one frequency to the total energy corresponding to the time-spectrum matrix to obtain energy proportion features, wherein the first preset peak value is lower than the second preset peak value; calculating the mean square error between the spectra of adjacent frames in the time-spectrum matrix to obtain time-domain variance features; wherein the multiple time-frequency features include time-frequency distribution features, time-spectrum entropy features, short-circuit frequency features, energy proportion features, and time-domain variance features.

[0072] In one alternative embodiment, to quantify the primary frequency components where energy is most concentrated during arc combustion and to reflect the dominant vibration mode during stable arc combustion, a weighted average frequency is calculated to reduce interference from background noise and secondary frequency components, providing a frequency distribution characteristic that represents the center of the overall energy distribution of the arc. Since this value should fluctuate within a specific range during normal welding, a shift in the frequency distribution characteristic indicates a change in arc length or an abnormality in gas protection.

[0073] During stable combustion, energy distribution is relatively uniform, resulting in a high entropy value. However, during splashing or unstable arcing, energy concentrates in certain frequency bands, leading to a decrease in entropy. To measure the uniformity of frequency distribution, the entropy values ​​of all frames are averaged to obtain global characteristics, i.e., spectral entropy characteristics, thereby quantifying the disorder or randomness of the arc combustion state. Statistical averaging reduces fluctuations caused by single-frame noise, reflecting the frequency distribution pattern.

[0074] To identify the time sequence of low-frequency energy peaks representing short-circuit transition events in the time spectrum, the reciprocal of the time interval between adjacent peaks is calculated as an estimate of the short-circuit transition frequency. This quantifies the periodicity of droplet short-circuit transition and reflects the frequency stability of metal transfer during welding. By tracking the time-domain distribution of specific low-frequency components, frequency domain information is mapped back to the time domain, capturing the rhythm of short-circuit events.

[0075] This is because high-frequency components mainly originate from transient physical phenomena such as molten droplet spatter and high-frequency arc oscillations. By extracting the total energy of the high-frequency band (e.g., greater than 500Hz) in the time spectrum and calculating the proportion of the high-frequency band to the total energy of the entire time spectrum, the intensity of high-frequency noise during welding can be quantified. Therefore, the energy proportion characteristic is an important feature for identifying spatter defects and assessing the intensity of the molten pool, and can effectively distinguish between normal welding and unstable states that produce a large amount of spatter.

[0076] Time-domain variance characteristics reflect the non-stationarity of the welding process. When welding parameters fluctuate, gas flow is unstable, or a slight short circuit exists, the spectral shape changes frequently and significantly, leading to an increase in the value of the time-domain variance characteristic. Therefore, time-domain variance characteristics are important features for assessing the dynamic stability of the welding process and identifying transient anomalies. By comparing the power spectrum differences between the m-th and (m+1)-th frames in the time-spectrum matrix at the same frequency point, and calculating the mean or variance of all inter-frame differences, the severity of the spectrum change over time can be measured. This quantifies the amplitude and frequency of dynamic fluctuations in the welding process, reflecting the speed and randomness of arc state changes. Thus, time-domain instability is transformed into frequency-domain statistics, revealing the dynamic characteristics of the welding process by capturing continuous changes in the spectral shape.

[0077] Optionally, the weld quality is inspected based on multiple time-frequency features to obtain inspection results, including: determining the detection threshold ranges corresponding to the multiple time-frequency features, wherein the time-frequency features and the detection threshold ranges correspond one-to-one; and obtaining the inspection results based on the detection threshold ranges corresponding to the multiple time-frequency features and the multiple time-frequency features.

[0078] The aforementioned detection threshold range refers to a normal fluctuation range established based on statistical analysis of the time-frequency characteristic data of historical qualified welds, used to determine whether multiple time-frequency characteristics are abnormal. The detection threshold range has online adaptive updating capabilities. It is not fixed but dynamically adjusted using a sliding window mechanism based on recent qualified weld data as welding conditions, plate thickness, welding speed, and other parameters change, ensuring that the detection always aligns with the current production status.

[0079] The detection threshold range is set based on statistical principles, assuming that the characteristic distribution of a normal welding process follows a certain statistical law, and anything exceeding this range is considered abnormal. By introducing an adaptive update mechanism, it can automatically compensate for reference drift caused by electrode wear, gas flow fluctuations, or environmental interference. The accuracy of the detection threshold range directly determines the false alarm rate and the missed alarm rate. A reasonable detection threshold range can effectively distinguish between normal process fluctuations and actual defects, providing a basis for graded quality judgment.

[0080] In one optional embodiment, in order to transform abstract time-frequency features into quality limits with clear physical meaning and statistical basis, and to ensure that each time-frequency feature has a corresponding judgment criterion, a mapping relationship between time-frequency features and detection threshold ranges can be established to provide a reference for subsequent quality judgment, thereby achieving multi-dimensional and refined monitoring of complex welding processes.

[0081] Then, the extracted time-frequency features are compared with their corresponding detection threshold ranges, and the quality level of the weld (such as qualified, warning, or unqualified) is determined by combining preset logic. By comprehensively considering the deviation of multiple time-frequency features, a comprehensive and objective assessment of the weld quality is made, avoiding missed detections or false alarms caused by anomalies in a single feature.

[0082] Optionally, based on the detection threshold ranges corresponding to multiple time-frequency features and the multiple time-frequency features, the detection result is obtained, including: based on the detection threshold ranges corresponding to multiple time-frequency features, determining whether the multiple time-frequency features are abnormal, and obtaining the number of abnormal time-frequency features, wherein the abnormal time-frequency features are not within the detection threshold range corresponding to the abnormal time-frequency features; and obtaining the detection result based on the number of abnormal time-frequency features.

[0083] In one optional embodiment, since different time-frequency features have different dimensions, physical meanings, and sensitivities to interference, anomaly identification of time-frequency features can be performed using detection threshold ranges, which can avoid misjudgment caused by dimension confusion.

[0084] After completing the anomaly assessment of all time-frequency features, the system does not directly use the anomaly of a single feature as the final conclusion. Instead, it accumulates and counts all anomaly labels to obtain the number of anomalies for each time-frequency feature. This process removes the specific deviation magnitude and dimensional information of each time-frequency feature, effectively compressing the data dimensionality. By aggregating multiple scattered anomaly labels into a statistically significant anomaly count, it provides a pure, dimensionless, count-based intermediate variable for subsequent decision-making.

[0085] Ultimately, the quality level of the weld is comprehensively assessed based on the number of anomalies. Relying on the collaborative consensus of multiple time-frequency characteristics improves the practicality and reliability of online detection.

[0086] Optionally, based on the number of anomalies with abnormal time-frequency characteristics, the detection results are obtained, including: if the number of anomalies is less than or equal to a first number, the quality level of the weld is determined to be level one; if the number of anomalies is greater than the first number but less than a second number, the quality level of the weld is determined to be level two, wherein the first number is less than the second number and the quality of level one is less than that of level two; if the number of anomalies is greater than or equal to the second number, the quality level of the weld is determined to be level three, wherein the quality of level two is less than that of level three.

[0087] In one optional embodiment, if the number of anomalies is less than or equal to a first number, the quality level of the weld can be determined to be the first level, meaning that the number of anomalies is at a low level (e.g., 0 or 1). The weld is then judged to be in a normal or slightly fluctuating state and assigned a high quality level. Minor, non-essential signal fluctuations are treated as normal process noise, avoiding false alarms caused by oversensitivity.

[0088] When the number of anomalies is greater than the first number but less than the second number, that is, when the anomalies are at an intermediate level (e.g., 2 or 3), it means that there are obvious anomalies but they have not yet constituted serious defects, thus identifying welds with potential quality risks but not yet completely failed.

[0089] If the number of abnormalities is greater than or equal to the second number, the quality level of the weld is determined to be the third level. That is, when the number of abnormalities is large, the weld can be judged to have a major quality defect and assigned the lowest quality level (such as C level unqualified) so as to intercept welds with serious defects. For example, by triggering a hard alarm signal, the welding operation can be terminated immediately or the weld can be marked as scrap, thereby avoiding the expansion of defects and causing greater material waste or safety hazards.

[0090] Optionally, the detection threshold ranges corresponding to multiple time-frequency features are determined, including: acquiring multiple historical time-frequency features of historical welds; for any one time-frequency feature, performing statistics on the historical time-frequency features corresponding to that one time-frequency feature to obtain the statistical value of that one time-frequency feature; and determining the detection threshold range of that one time-frequency feature based on the statistical value.

[0091] In one optional embodiment, time-frequency characteristic data of qualified welds that have been manually verified or system-determined in the past are collected and stored to establish a benchmark dataset representing "normal welding conditions," ensuring the representativeness of subsequent threshold settings and avoiding statistical bias caused by single or small samples. By accumulating diverse qualified process data (covering different plate thicknesses, welding speeds, gas ratios, and other operating conditions), the sample distribution range is enriched, enabling the detection threshold range to reflect the true statistical characteristics of the welding process under normal fluctuations.

[0092] Then, for any time-frequency feature, statistical analysis is performed using historical time-frequency features to obtain the statistical value of any time-frequency feature, such as mean, standard deviation, or variance. This allows for the quantification of the range and fluctuation pattern of the time-frequency feature during the normal welding process, revealing the statistical distribution characteristics of different feature dimensions.

[0093] Therefore, based on statistical values, the detection threshold range of any time-frequency feature can be determined, and the statistical results can be transformed into executable control rules, enabling the rapid and accurate determination of whether the time-frequency feature is abnormal according to the detection threshold range.

[0094] Optionally, the electrical signal includes a current signal; based on the signal values ​​of multiple electrical signals, the multiple electrical signals are divided into sub-signal segments of different time periods, including: a first acquisition time for detecting the current signal rising from its initial value to the arc ignition detection threshold; a second acquisition time for acquiring the current signal rising from the arc ignition detection threshold and stabilizing at a preset state, wherein the preset state is that the signal value of the current signal is within a preset range, and the duration of the signal value of the current signal being within the preset range reaches a preset time, and the second acquisition time is later than the first acquisition time; a third acquisition time for detecting the current signal value being lower than the arc extinction detection threshold when the electrical signal is stable at the preset state, wherein the third acquisition time is later than the second acquisition time; a fourth acquisition time for determining the current signal decreasing from the arc extinction detection threshold to its initial value, wherein the fourth acquisition time is later than the third acquisition time; based on the first acquisition time, the second acquisition time, the third acquisition time, and the fourth acquisition time, the multiple electrical signals are segmented to obtain sub-signal segments of different time periods.

[0095] In one optional embodiment, the first acquisition time is used to detect the current signal rising from its initial value to the arc-ignition detection threshold. That is, the trajectory of the current signal over time is monitored to identify the specific moment when the current starts rising from zero or a low level and first crosses a preset arc-ignition threshold (e.g., 20% of the steady-state current average). By capturing the rising edge characteristics of the current signal, positioning errors caused by manual operation delays or sensor response times are reduced, ensuring the integrity of the arc-ignition phase.

[0096] Then, the second acquisition time is used to acquire the current signal as it rises from the arc detection threshold and stabilizes at a preset state. The system monitors whether the current signal, after exceeding the arc detection threshold, enters a state with relatively small current fluctuations, within a specific range near the set target value (e.g., ±15% of the steady-state current average), and whether this state lasts for more than a minimum time window (e.g., 50ms). This confirms that the welding process has entered a stable arc-burning stage, and the moment this stable state is established is recorded. This allows the second acquisition time to distinguish between the arc-ignition transition stage and the stable combustion stage, eliminating unstable fluctuations at the moment of arc ignition, ensuring that feature extraction is primarily based on a stable physical process, effectively filtering out transient jitter and false signals during arc ignition, and accurately defining the start boundary of the stable arc-burning stage. This helps to segment the welding process stages, ensuring that subsequent time-frequency feature extraction mainly comes from stable and representative stages of arc combustion, thereby improving the signal-to-noise ratio and effectiveness of welding quality judgment.

[0097] The system monitors the current signal, which is already in a stable state. When a decrease in the amplitude of the current signal is detected and it first falls below the preset arc extinguishing threshold (e.g., 10% of the average steady-state current), this moment is recorded as the third acquisition time. This captures the turning point in the welding process from steady state to completion, i.e., the moment when the arc length begins to lengthen or the current begins to decrease rapidly in preparation for arc extinguishing. By identifying the falling edge characteristics of the current signal, the end point of the stable arc-burning stage is determined, thereby defining the right boundary of the stable combustion range and preventing the unstable signals from the initial stage of arc extinguishing from being mixed into the time-frequency feature analysis.

[0098] Furthermore, the fourth acquisition time is determined as the current signal decreases from the arc extinguishing detection threshold to its initial value. This fourth acquisition time is the moment when the monitored current signal continues to decrease after falling below the arc extinguishing threshold until it returns to the initial background level of zero or near zero. This fourth acquisition time marks the complete end of the welding process, i.e., the instant the arc is extinguished and the current is cut off. By capturing the zero-return characteristic of the current signal, the endpoint of the entire welding time series is determined.

[0099] Therefore, based on the first acquisition time, the second acquisition time, the third acquisition time, and the fourth acquisition time, multiple electrical signals can be segmented into three independent sub-signal segments. This allows for the implementation of a segmented signal processing strategy, enabling the extraction of targeted time-frequency features for different physical stages of the welding process that have different typical defect modes and signal characteristics.

[0100] Optionally, the above method further includes: counting the number of welds that have been detected; when the number of welds that have been detected reaches the update number, obtaining a preset number of target welds from the welds that have been detected based on the quality level; and obtaining a new detection threshold range based on the time-frequency characteristics of the target welds.

[0101] In one optional embodiment, the number of welds that have been inspected is counted. By recording and accumulating the total number of welds that have completed the inspection process, the operating status and data sampling progress of the weld station are monitored. This provides the time or quantity triggering conditions for the subsequent threshold update mechanism, ensuring that the threshold update is carried out after a certain amount of new data has been accumulated, and avoiding oscillations caused by frequent updates.

[0102] Then, when the update step size is reached, qualified welds are selected, and the most recent continuous data (such as the most recent N data) is selected as the time-frequency feature used for this update. This ensures that the samples used to construct the new threshold represent the normal welding process state. At the same time, by selecting the welds with the most recent time, the current working condition characteristics are reflected, thus constructing a high-quality and timely update basis.

[0103] Then, using the time-frequency characteristics of the selected target welds, a new detection threshold range corresponding to each feature dimension is recalculated. This allows the new detection threshold range to dynamically adapt to changes during the welding process, such as electrode wear, base material differences, or parameter fine-tuning. The detection threshold range is iterated by automatically adjusting the judgment boundary.

[0104] Optionally, the above method further includes: associating and storing the detection results, target signal, and multiple time-frequency features with the vehicle identification and weld code of the weld, respectively.

[0105] In one optional embodiment, the detection results, target signals, and multiple time-frequency features are linked one-to-one with specific vehicle identifiers and specific weld codes (such as location numbers or function numbers), and then persistently stored in a database. This constructs a complete digital archive of welding quality, achieves data mapping, and ensures that the quality information of each weld is verifiable and traceable, facilitating subsequent data retrieval, statistical analysis, and traceability, thus providing data support for weld quality management.

[0106] Optionally, the above method also includes: controlling the welding station based on the test results.

[0107] In one optional embodiment, based on the detection results, corresponding control commands can be sent to the welding robot controller or production execution system to achieve closed-loop linkage between the detection system and the execution system. This directly translates the quality inspection results into real-time intervention or feedback for the welding process, ensuring that abnormal states are handled promptly and preventing defects from escalating or defective products from flowing into the next process. Through automated control logic, adaptive adjustment and safety protection of the welding process are achieved. For example, automatically reducing the welding speed or adjusting wire feeding parameters when severe spatter or arc instability is detected, or forcibly shutting down the machine in case of a serious malfunction to protect the equipment and workpiece. This constructs an integrated "detection-control" intelligent welding system, improving the automation level and response speed of the welding process.

[0108] Optionally, based on the detection results, the welding station can be controlled, including: if the quality level is less than or equal to the second level, controlling the welding station to continue welding the next weld; if the quality level is greater than the second level, outputting an early warning message and controlling the welding station to stop welding.

[0109] In one optional embodiment, when the quality level is less than or equal to the second level, it means the welding process is under control. A command to continue operation can be sent to the welding robot or controller, seamlessly connecting to the next weld seam. This ensures that the continuity and efficiency of the production process are not affected when the welding quality is within the normal or slightly fluctuating range, avoiding unnecessary downtime due to overly sensitive alarms and thus maintaining production stability. Logical judgment filters out conditions that allow continued production, achieving automated flow of the normal production process and reducing the need for manual intervention.

[0110] If the quality level is higher than the second level, it means the weld quality has failed inspection. An early warning message can be issued, and welding at the workstation can be stopped. For example, a red alarm message can be displayed on the human-machine interface to alert the operator. Simultaneously, an emergency stop command is sent to the welding control system to interrupt the current welding operation or suspend subsequent welding processes. This allows for the prevention of major quality defects at their source, preventing defective products from flowing into the next process or the final product. The alarm also alerts personnel to investigate the cause of the malfunction. This rigid control measure achieves immediate isolation of quality risks, prevents defects from escalating, and buys time for subsequent equipment debugging or process parameter adjustments.

[0111] According to an embodiment of this application, an embodiment of a weld detection circuit is provided.

[0112] This embodiment provides a weld detection circuit. Figure 2 This is a schematic diagram of a weld detection circuit according to an embodiment of this application, as shown below. Figure 2 As shown, the circuit includes:

[0113] The acquisition circuit 20 is connected to the weld seam S400 at the welding station and is used to acquire multiple electrical signals in the welding circuit where the weld seam is located during the welding process at the welding station.

[0114] The controller 22, connected to the acquisition circuit, is used to divide multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of multiple electrical signals, extract time-frequency features of target signal segments in sub-signal segments of different time periods to obtain multiple time-frequency features of the weld, and perform quality inspection on the weld based on the multiple time-frequency features to obtain the inspection result. The signal values ​​of multiple target signals in the target signal segment are within a preset range, and the duration of the signal values ​​of multiple target signals being within the preset range reaches a preset time. The inspection result includes the quality level of the weld.

[0115] The detection circuit in this embodiment includes a data acquisition circuit and a controller. The data acquisition circuit can capture the electrical signals flowing through the welding circuit in real time during the welding process, providing raw data for subsequent digital signal processing and quality analysis. It can acquire electrical signals reflecting key physical information such as welding arc stability and droplet transfer state in real time without damaging the weld.

[0116] The controller connects to the acquisition circuit. At the software or algorithm level, a microprocessor or embedded system is used to segment and process multiple electrical signals. Time-frequency analysis algorithms are applied to representative target signal segments to extract multi-dimensional time-frequency features, overcoming the limitations of single time-domain or frequency-domain analysis and comprehensively capturing the dynamic characteristics of the welding process. This provides rich and reliable data support for subsequent quality assessment, thereby determining the quality level of the weld and achieving automated and graded evaluation of welding quality.

[0117] Optionally, the acquisition circuit includes: a sensor connected to the weld seam for acquiring multiple analog signals in the welding circuit; a filter located between the sensor and the data acquisition card for filtering the multiple analog signals to obtain multiple processed analog signals; and a data acquisition card located between the filter and the controller for converting the multiple processed analog signals into multiple electrical signals.

[0118] In this embodiment, the acquisition circuit deploys sensors at the welding power supply output or in the circuit to capture instantaneous physical quantities of current and voltage during the welding process in real time and convert them into proportional analog signals. This non-invasive acquisition method ensures accurate electrical signals can be obtained even in environments with strong electromagnetic interference.

[0119] Then, using a filter between the sensor and the data acquisition card, multiple analog signals are filtered to reduce high-frequency noise and interference above the Nyquist frequency, as well as uncorrelated high-frequency components introduced by the sensor, resulting in multiple processed analog signals. This prevents high-frequency noise from causing spectral aliasing during sampling, which would lead to digital signal distortion. Simultaneously, it suppresses high-frequency electromagnetic interference generated during welding, protects subsequent circuits, improves the signal-to-noise ratio, and avoids feature extraction errors caused by aliasing effects.

[0120] The data acquisition card in the acquisition circuit, located between the filter and the controller, can convert multiple processed analog signals into multiple electrical signals. For example, a high-precision analog-to-digital converter can be used to synchronously sample and quantize multiple processed analog signals at a set sampling rate (e.g., 10kHz) and resolution, converting them into a sequence of electrical signals so that the controller can perform subsequent digital signal processing algorithms. Through the conversion from the analog domain to the digital domain, continuous analog signals are transformed into discrete data streams that can be stored and processed by computers or embedded systems.

[0121] Optionally, the sensors include: a voltage sensor connected in parallel with the weld seam for acquiring multiple voltage signals; and a current sensor connected in series with the weld seam for acquiring multiple current signals; wherein the multiple analog signals include multiple voltage signals and multiple current signals.

[0122] Because arc voltage is directly related to arc length, droplet transfer mode, and the stability of welding energy input, fluctuations in the voltage signal can sensitively reflect transition processes such as short circuits and arcing. Voltage sensors can be connected in parallel with the weld to collect multiple voltage signals. For example, a high-impedance differential voltage sensor can be connected in parallel between the tungsten electrode (or welding torch) and the workpiece in the welding circuit to directly measure the potential difference across the arc, thereby obtaining voltage signals reflecting the arc combustion state and arc length changes. Through non-contact or high-impedance parallel connection, the instantaneous value of the voltage waveform can be accurately extracted without affecting the current in the welding circuit, providing voltage-dimensional data support for subsequent analysis of arc stability.

[0123] The current signal determines the generation of welding heat and the formation of the molten pool. Its waveform characteristics (such as peak current, short-circuit peak current, and base current) are key indicators for judging welding defects (such as lack of fusion and burn-through). A current sensor can be connected in series with the weld to collect multiple current signals. For example, a current sensor can be connected in series in the output circuit of the welding power source, allowing all welding current to flow through it, thereby acquiring current signals reflecting the welding heat input, droplet transition frequency, and short-circuit current magnitude. Real-time monitoring of the welding circuit current through this series connection ensures that the collected current signal accurately reflects the actual load of the welding power source, providing current-dimensional data support for time-frequency feature extraction. Parallel acquisition of voltage and current signals can more comprehensively capture the physical phenomena of the welding process, improving the richness of time-frequency feature description and the accuracy of quality judgment. Combining both methods enables more comprehensive and robust monitoring of the welding process.

[0124] According to an embodiment of this application, an embodiment of a weld inspection system is provided.

[0125] This embodiment provides a weld inspection system. Figure 3This is a schematic diagram of a weld inspection system according to an embodiment of this application, as shown below. Figure 3 As shown, the system includes:

[0126] The signal acquisition module 30 is used to acquire multiple electrical signals in the welding circuit where the weld is located during the welding process of the weld at the welding station.

[0127] The feature transformation module 32 is used to divide multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of multiple electrical signals, extract time-frequency features of target signal segments in sub-signal segments of different time periods, and obtain multiple time-frequency features of the weld. Among them, the signal values ​​of multiple target signals in the target signal segments are within a preset range, and the duration of the signal values ​​of multiple target signals being within the preset range reaches a preset time.

[0128] The quality assessment module 34 is used to perform quality inspection on the weld based on multiple time-frequency features and obtain the inspection results, including the quality level of the weld.

[0129] The signal acquisition module of this detection system acquires multiple electrical signals generated during the welding process in real time through a hardware interface, thus enabling complete and real-time capture of physical quantities reflecting the welding arc state, droplet transfer, and changes in heat input. The feature transformation module preprocesses and segments the multiple electrical signals, identifying the electrical signal during the stable arc-burning stage as the target signal segment. Then, a time-frequency analysis algorithm is applied to this target signal segment to extract multi-dimensional time-frequency features to reflect the welding state of the weld. Based on the extracted time-frequency features, the quality assessment module achieves automated and graded evaluation of welding quality, improving the accuracy and robustness of the detection.

[0130] The technical solution proposed in this application is described below with reference to an optional embodiment. This application proposes an online detection method for electrical signals in the welding process based on the combination of short-time fourier transform (STFT) time-frequency feature extraction and adaptive dynamic threshold quality control. It is applicable to the real-time weld quality judgment scenario of gas metal arc welding (MIG welding) of aluminum alloy car bodies.

[0131] Specifically, the detection method involves steps such as electrical signal acquisition and preprocessing, automatic segmentation of the welding stage, extraction of STFT time-frequency features, establishment and updating of adaptive dynamic threshold, and evaluation of weld quality grading.

[0132] In the electrical signal acquisition and preprocessing step, the current signal I(t) and voltage signal U(t) during the welding process are simultaneously acquired using a high-precision current sensor (sampling rate ≥10kHz) and a voltage sensor. Adaptive median filtering preprocessing is performed on the current signal I(t) and voltage signal U(t) to remove random pulse noise while preserving the transient change characteristics of the signal.

[0133] The welding process is automatically segmented based on the amplitude characteristics of the current signal. It automatically identifies and segments the welding process into three stages: the arc ignition stage (current rises from zero to a stable value), the stable arc burning stage (current fluctuates around the set value), and the arc extinguishing stage (current drops from a stable value to zero).

[0134] In the STFT time-frequency feature extraction step, the current signal I(t) and voltage signal U(t) during the stable arc burning stage are subjected to short-time Fourier transforms to obtain time-frequency spectrum matrices SI(t,f) and SU(t,f). If a Hanning window is selected with a window length L (covering approximately 5-10 short-circuit transition cycles) and a frame shift step size of L / 4, a multi-dimensional time-frequency feature vector F is extracted from the time-frequency spectrum matrix. F = [F1, F2, F3, F4, F5]. Here, F1 is the time-frequency energy centroid frequency, which can reflect the dominant frequency characteristics of arc combustion through weighted average frequency. F2 is the time-frequency spectrum entropy, which reflects the uniformity of frequency distribution and characterizes arc stability. F3 is the short-circuit transition frequency, which can be estimated by the time-domain interval of low-frequency peaks in the current signal time-frequency spectrum, reflecting the stability of the short-circuit transition cycle. F4 is the high-frequency energy proportion, which reflects the degree of spatter by the ratio of high-frequency energy to total energy. F5 is the time-domain variance of the time-frequency spectrum, which reflects the dynamic fluctuations of the welding process by the degree of difference between the spectra of each frame. By using multidimensional time-frequency eigenvectors, the time-domain local characteristics and frequency-domain energy distribution characteristics of the welding process are comprehensively characterized.

[0135] In the adaptive dynamic threshold establishment and update step, initialization is first performed. An initial normal feature distribution model is established using the time-frequency feature vectors of the first batch of N0 (N0≥30) qualified welds. The mean μ0 and standard deviation σ0 of each feature dimension are calculated, and the initial threshold interval [μ0-kσ0, μ0+kσ0] is set (k is a control coefficient, and the initial value can be 3). Then, online updates are performed. After every M welds are inspected (M is the update step size, which is configurable), the mean μ and standard deviation σ are updated using the feature data of the most recent N qualified welds (sliding window), realizing dynamic adaptive adjustment of the threshold. If the failure rate of a batch exceeds 10%, the threshold update is paused and manual review is triggered.

[0136] Then, anomaly detection and judgment are performed. When any dimension of the time-frequency feature vector of a weld exceeds the corresponding threshold range, an anomaly warning is triggered. When the number of feature dimensions exceeding the threshold is ≥2, the weld is determined to be unqualified, a red alarm can be issued, and a PLC (Programmable Logic Controller) shutdown signal can be output.

[0137] In the weld quality grading and evaluation step, the weld quality is evaluated in three levels based on the deviation of the time-frequency feature vector from the threshold range: if all feature dimensions are within the threshold range, the weld is qualified. If one feature dimension exceeds the threshold range, the weld may have a minor defect, triggering an early warning. If more than two feature dimensions exceed the threshold range, the weld is deemed unqualified, triggering an alarm and stopping the machine.

[0138] Therefore, STFT time-frequency analysis simultaneously captures temporal local characteristics (short-circuit transition, spatter events) and frequency-domain energy distribution (arc stability). The 10-dimensional feature vector improves the detection sensitivity for welding defects compared to a single feature quantity. The sliding window dynamic threshold update mechanism eliminates the need for manual recalibration when switching welding parameters, automatically adapting to changes in plate thickness, welding speed, shielding gas ratio, and other operating conditions. The three-level evaluation system provides process engineers with more refined quality information, supporting continuous optimization of welding parameters. Furthermore, STFT avoids the end-point effects and modal mixing problems of EMD, and its algorithm output is stable, making it suitable for large-scale online industrial inspection. STFT has low computational complexity and can run efficiently on industrial edge computing devices.

[0139] This embodiment also provides an online detection system for welding quality electrical signals, including: a signal acquisition module, a stage segmentation module, an STFT time-frequency feature extraction module, an adaptive threshold management module, a quality grading evaluation module, and a result display and storage module.

[0140] The inspection process is explained using online inspection of weld quality in MIG welding of aluminum alloys as an example.

[0141] At the welding robot station, a Hall effect current sensor (range 0~500A, sampling rate 10kHz, accuracy ±0.5%) is connected in series at the welding power supply output, and a differential voltage sensor (range 0~60V, sampling rate 10kHz) is connected in parallel. The output signals of the two sensors are passed through an anti-aliasing low-pass filter (cutoff frequency 4kHz), and then the current signal I(t) and voltage signal U(t) are synchronously acquired by a synchronous data acquisition card (16-bit ADC) at a sampling rate of 10kHz.

[0142] Then, adaptive median filtering is performed on the original current signal I(t) and voltage signal U(t) respectively. The filtering window length is adaptively adjusted according to the local variance of the signal. If the local variance is large, the window is shortened to preserve transient characteristics, and if the local variance is small, the window is extended to effectively remove noise. The window length range is 3~15 sampling points (corresponding to 0.3ms~1.5ms).

[0143] Furthermore, the welding stage is automatically segmented. For example, based on the amplitude characteristics of the current signal I(t), an arc ignition detection threshold Istart (which can be 20% of the steady-state current average) and an arc extinguishing detection threshold Iend (which can be 10% of the steady-state current average) can be set. The arc ignition stage is defined as the time t1 when the current signal rises from 0 and first exceeds Istart, until t2 when the current signal stabilizes within ±15% of the steady-state current average for more than 50 ms. The stable arc ignition stage is defined as the time from t2 to t3 when I(t) first falls below Iend. The arc extinguishing stage is defined as the time from t3 to t4 when I(t) drops to 0.

[0144] STFT time-frequency feature extraction mainly targets the stable arc burning stage [t2,t3].

[0145] The current signal I(t) during the stable arcing phase is processed by STFT. A Hanning window is selected with a window length of L = 512 points (corresponding to 51.2 ms, covering approximately 5 to 8 MIG welding short-circuit transition cycles). The frame shift step size is L / = 128 points (corresponding to 12.8 ms), and the number of FFT points is NFFT = 512, thus obtaining the time-frequency spectrum matrix SI(m,k), where m is the frame index and k is the frequency bin index (frequency resolution 19.5 Hz).

[0146] Five-dimensional time-frequency features are extracted from SI(m,k), such as F1 (time-frequency energy centroid frequency), which is the average of all frames. For example, the F1 range for a qualified weld bead is 80~150Hz (corresponding to the short-circuit transition frequency range of MIG welding). F2 (time-frequency spectral entropy), calculated by averaging the normalized power spectrum of each frame, is used to calculate the Shannon entropy for each frame. The time-frequency spectral entropy of a qualified weld bead should be at a high level (uniform frequency components when the arc is stable), and the entropy value decreases significantly when spatter is severe. F3 (short-circuit transition frequency) is calculated by taking the time sequence of moments when the current value in the detected current signal is lower than the short-circuit detection threshold (40% of the steady-state average), calculating the time interval between adjacent short-circuit events, and taking the reciprocal of the average as the estimated short-circuit transition frequency. F4 (high-frequency energy proportion) is the average of all frames; the proportion of high-frequency energy increases when spatter is severe. F5 (time-frequency spectral variance) is calculated by calculating the root mean square error between the spectra of adjacent frames, reflecting the dynamic fluctuations of the welding process. The F5 value increases when the welding is unstable.

[0147] The same 5-dimensional features are extracted from the voltage signal U(t) to obtain a total of 10-dimensional feature vector F=[F1I,F2I,F3I,F4I,F5I,F1U,F2U,FU,FU,F5U].

[0148] Then, the weld quality is graded, and the number of feature dimensions (nout) exceeding the threshold range in the current weld feature vector is counted. If nout=0, it is judged as Grade A (qualified), the result is stored, and the process continues to the next weld. If nout=10, it is judged as Grade B (warning), a yellow warning message is displayed on the monitoring screen, the warning information is recorded, and welding continues. If nout≥2: it is judged as Grade C (unqualified), a red alarm is triggered, an alarm signal is output to the PLC, the current welding station operation is stopped, and manual intervention is required.

[0149] Finally, the results are bound and stored. For example, the weld quality rating results, 10-dimensional feature vector values, and original electrical signal fragments are bound with the vehicle identification number and weld number and stored in the quality database to support subsequent quality traceability and welding parameter optimization analysis.

[0150] like Figure 4 An alternative detection method is demonstrated. This method includes acquiring electrical signals, performing time-frequency transformation, feature extraction, feature fusion, and adaptive threshold calculation to obtain a judgment threshold. After threshold comparison, the quality level is output.

[0151] like Figure 5 An optional detection circuit is shown. The detection circuit includes a welding power source S100, a weld seam S400, a voltage sensor S200, a current sensor S300, a filter circuit S600, a data acquisition card S700, and a controller S800.

[0152] like Figure 6 An optional detection system is demonstrated. This system includes a signal acquisition module, a time-frequency transformation module, a feature extraction module, a feature fusion module, a threshold adaptive module, a quality judgment module, a communication interface module, a data storage module, a display module, and an alarm module.

[0153] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0154] According to an embodiment of this application, an embodiment of a weld inspection device is provided. It should be noted that the device can be used to perform the above-described weld inspection method.

[0155] This embodiment provides a weld inspection device. Figure 7 This is a schematic diagram of a weld inspection device according to an embodiment of this application, as shown below. Figure 7 As shown, the device includes:

[0156] The acquisition module 70 is used to acquire multiple electrical signals in the welding circuit where the weld is located during the welding process of the weld at the welding station.

[0157] The segmentation module 72 is used to divide multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of multiple electrical signals.

[0158] Extraction module 74 is used to extract time-frequency features from target signal segments in sub-signal segments of different time periods to obtain multiple time-frequency features of the weld, wherein the signal values ​​of multiple target signals in the target signal segment are within a preset range, and the duration of the signal values ​​of multiple target signals being within the preset range reaches a preset time.

[0159] The detection module 76 is used to perform quality inspection on the weld based on multiple time-frequency features and obtain the detection results, including the quality level of the weld.

[0160] Optionally, the extraction module is also used to: filter multiple target signals in the target signal segment to obtain multiple transient features of the multiple target signals; perform short-time Fourier transform on the multiple transient features of the multiple target signals to obtain the time-frequency matrix of the target signal segment; and perform time-frequency statistical calculations based on the time-frequency matrix to obtain multiple time-frequency features.

[0161] Optionally, the extraction module is further configured to: perform weighted calculations on the frequencies of each frame in the time-spectrum matrix to obtain frequency distribution features; calculate the mean of the time-spectrum entropy of each frame in the time-spectrum matrix to obtain time-spectrum entropy features; detect time series of frequencies in the time-spectrum matrix that are lower than a first preset peak value, and obtain short-circuit frequency features based on the time series; detect at least one frequency in the time-spectrum matrix that is higher than a second preset peak value, and calculate the ratio of the energy of at least one frequency to the total energy corresponding to the time-spectrum matrix to obtain energy proportion features, wherein the first preset peak value is lower than the second preset peak value; calculate the mean square error between the spectra of adjacent frames in the time-spectrum matrix to obtain time-domain variance features; wherein the multiple time-frequency features include time-frequency distribution features, time-spectrum entropy features, short-circuit frequency features, energy proportion features, and time-domain variance features.

[0162] Optionally, the detection module is also used to: determine the detection threshold ranges corresponding to multiple time-frequency features, wherein the time-frequency features and the detection threshold ranges correspond one-to-one; and obtain the detection results based on the detection threshold ranges corresponding to the multiple time-frequency features and the multiple time-frequency features.

[0163] Optionally, the detection module is further configured to: determine whether multiple time-frequency features are abnormal based on the detection threshold ranges corresponding to the multiple time-frequency features respectively, and obtain the number of abnormal time-frequency features, wherein the abnormal time-frequency features are not within the detection threshold range corresponding to the abnormal time-frequency features; and obtain the detection result based on the number of abnormal time-frequency features.

[0164] Optionally, the detection module is further configured to: determine the quality level of the weld as a first level when the number of abnormalities is less than or equal to a first quantity; determine the quality level of the weld as a second level when the number of abnormalities is greater than the first quantity and less than a second quantity, wherein the first quantity is less than the second quantity and the quality of the first level is less than that of the second level; and determine the quality level of the weld as a third level when the number of abnormalities is greater than or equal to the second quantity, wherein the quality of the second level is less than that of the third level.

[0165] Optionally, the detection module is also used to: acquire multiple historical time-frequency features of historical welds; for any given time-frequency feature, perform statistics on the historical time-frequency features corresponding to that time-frequency feature to obtain a statistical value for that time-frequency feature; and determine the detection threshold range for any given time-frequency feature based on the statistical value.

[0166] Optionally, the electrical signal includes a current signal; the segmentation module is further configured to: detect a first acquisition time when the current signal rises from its initial value to the arc ignition detection threshold; acquire a second acquisition time when the current signal rises from the arc ignition detection threshold and stabilizes at a preset state, wherein the preset state is that the signal value of the current signal is within a preset range, and the duration of the signal value of the current signal being within the preset range reaches a preset time, and the second acquisition time is later than the first acquisition time; detect a third acquisition time when the signal value of the current signal is lower than the arc extinction detection threshold when the electrical signal is stable at the preset state, wherein the third acquisition time is later than the second acquisition time; determine a fourth acquisition time when the current signal decreases from the arc extinction detection threshold to its initial value, wherein the fourth acquisition time is later than the third acquisition time; and segment multiple electrical signals based on the first acquisition time, the second acquisition time, the third acquisition time, and the fourth acquisition time to obtain sub-signal segments of different time periods.

[0167] Optionally, the above-mentioned device further includes an update module, used for: counting the number of welds that have been detected; when the number of welds that have been detected reaches the update quantity, obtaining a preset number of target welds from the welds that have been detected based on the quality level; and obtaining a new detection threshold range based on the time-frequency characteristics of the target welds.

[0168] Optionally, the above device further includes a storage module for:

[0169] The detection results, target signals, and multiple time-frequency features are associated and stored with the vehicle identification and weld code of the weld, respectively.

[0170] Optionally, the above device also includes a control module for controlling the welding station based on the detection results.

[0171] Optionally, the control module is also used to: control the welding station to continue welding the next weld when the quality level is less than or equal to the second level; and output a warning message and control the welding station to stop welding when the quality level is greater than the second level.

[0172] Embodiments of this application also provide a vehicle, including: a body, wherein welds on the body are detected by the detection methods in various embodiments of this application, or by the detection circuits in various embodiments of this application, or by the detection systems in various embodiments of this application.

[0173] This application also provides an electronic device, including a memory and a processor, wherein the memory is used to store computer programs; and the processor is used to execute the programs stored in the memory to implement the methods in various embodiments of this application.

[0174] Embodiments of this application also provide a computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the computer-readable storage medium is located to perform the methods of various embodiments of this application.

[0175] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the methods of various embodiments of this application.

[0176] Embodiments of this application also provide a computer program product, including a non-volatile computer-readable storage medium for storing a computer program that, when executed by a processor, implements the methods in various embodiments of this application.

[0177] Embodiments of this application also provide a computer program that, when executed by a processor, implements the methods described in the various embodiments of this application.

[0178] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0179] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0180] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0181] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0182] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0183] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for inspecting welds, characterized in that, include: During the welding process of the weld seam at the welding station, multiple electrical signals in the welding circuit where the weld seam is located are collected; Based on the signal values ​​of the multiple electrical signals, the multiple electrical signals are divided into sub-signal segments of different time periods; Time-frequency features are extracted from the target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld. The signal values ​​of multiple target signals in the target signal segments are within a preset range, and the duration for which the signal values ​​of multiple target signals are within the preset range reaches a preset time. Based on the aforementioned multiple time-frequency characteristics, the weld is subjected to quality inspection to obtain inspection results, wherein the inspection results include the quality grade of the weld.

2. The method according to claim 1, characterized in that, Time-frequency features are extracted from the target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld, including: Multiple target signals in the target signal segment are filtered to obtain multiple transient features of the multiple target signals; Short-time Fourier transform is performed on multiple transient features of the multiple target signals to obtain the time-spectrum matrix of the target signal segment; Based on the time-frequency spectrum matrix, time-frequency statistical calculations are performed to obtain the multiple time-frequency features.

3. The method according to claim 2, characterized in that, Based on the time-frequency spectrum matrix, time-frequency statistical calculations are performed to obtain the multiple time-frequency features, including: The frequency distribution characteristics are obtained by weighting the frequencies of each frame in the time-spectrum matrix. The mean of the temporal spectral entropy of each frame in the temporal spectral matrix is ​​calculated to obtain the temporal spectral entropy feature; The time series of frequencies below a first preset peak value in the time-spectrum matrix are detected, and short-circuit frequency characteristics are obtained based on the time series. Detect at least one frequency in the time-spectrum matrix that is higher than a second preset peak value, and calculate the ratio of the energy of the at least one frequency to the total energy corresponding to the time-spectrum matrix to obtain the energy proportion feature, wherein the first preset peak value is lower than the second preset peak value; The mean square error between the spectra of adjacent frames in the time-spectrum matrix is ​​calculated to obtain the time-domain variance characteristics; The multiple time-frequency features include time-frequency distribution features, time-spectrum entropy features, short-circuit frequency features, energy proportion features, and time-domain variance features.

4. The method according to claim 1, characterized in that, Based on the aforementioned multiple time-frequency characteristics, the weld seam is subjected to quality inspection to obtain inspection results, including: Determine the detection threshold range corresponding to each of the plurality of time-frequency features, wherein the time-frequency features and the detection threshold ranges correspond one-to-one; The detection result is obtained based on the detection threshold range corresponding to the multiple time-frequency features and the multiple time-frequency features respectively.

5. The method according to claim 4, characterized in that, Based on the detection threshold ranges corresponding to the multiple time-frequency features and the multiple time-frequency features, the detection result is obtained, including: Based on the detection threshold ranges corresponding to the multiple time-frequency features, it is determined whether the multiple time-frequency features are abnormal, and the number of abnormal time-frequency features is obtained, wherein the abnormal time-frequency features are not within the detection threshold range corresponding to the abnormal time-frequency features; The detection result is obtained based on the number of anomalies in the aforementioned abnormal time-frequency characteristics.

6. The method according to claim 5, characterized in that, The detection result is obtained based on the number of anomalies in the aforementioned abnormal time-frequency features, including: If the number of anomalies is less than or equal to the first number, the quality level of the weld is determined to be the first level; If the number of abnormalities is greater than the first number but less than the second number, the quality level of the weld is determined to be the second level, wherein the first number is less than the second number, and the quality of the first level is less than that of the second level. If the number of abnormalities is greater than or equal to the second number, the quality level of the weld is determined to be the third level, wherein the quality of the second level is lower than that of the third level.

7. The method according to claim 6, characterized in that, Determining the detection threshold ranges corresponding to the plurality of time-frequency features includes: Obtain multiple historical time-frequency features of historical welds; For any given time-frequency feature, the historical time-frequency features corresponding to that time-frequency feature are statistically analyzed to obtain the statistical value of that time-frequency feature. Based on the statistical values, the detection threshold range for any time-frequency feature is determined.

8. The method according to claim 1, characterized in that, The electrical signal includes a current signal; Based on the signal values ​​of the multiple electrical signals, the multiple electrical signals are divided into sub-signal segments of different time periods, including: The first acquisition time is used to detect the current signal rising from its initial value to the arc detection threshold. The second acquisition time is obtained when the current signal rises from the arc detection threshold and stabilizes in a preset state, wherein the preset state is when the signal value of the current signal is within a preset range and the duration of the signal value of the current signal being within the preset range reaches a preset time, and the second acquisition time is later than the first acquisition time. When the electrical signal is stable in the preset state, a third acquisition time is detected when the signal value of the current signal is lower than the arc extinction detection threshold, wherein the third acquisition time is later than the second acquisition time; Determine the fourth acquisition time when the current signal decreases from the arc extinction detection threshold to the initial value, wherein the fourth acquisition time is later than the third acquisition time; Based on the first acquisition time, the second acquisition time, the third acquisition time, and the fourth acquisition time, the multiple electrical signals are segmented to obtain sub-signal segments of different time periods.

9. The method according to any one of claims 1-8, characterized in that, The method further includes: Count the number of welds that have been inspected; If the number of welds that have been inspected reaches the update quantity, a preset number of target welds are obtained from the welds that have been inspected, based on the quality level. Based on the time-frequency characteristics of the target weld, a new detection threshold range is obtained.

10. The method according to any one of claims 1-8, characterized in that, The method further includes: The detection results, the target signal, and the multiple time-frequency features are associated and stored with the vehicle identification and the weld code of the weld, respectively.

11. The method according to any one of claims 1-8, characterized in that, The method further includes: Based on the test results, the welding station is controlled.

12. The method according to claim 11, characterized in that, Based on the detection results, the welding station is controlled, including: If the quality level is less than or equal to the second level, the welding station is controlled to continue welding the next weld. If the quality level is higher than the second level, an early warning message is output, and the welding station is controlled to stop welding.

13. A weld detection circuit, characterized in that, include: The acquisition circuit is connected to the weld seam at the welding station and is used to acquire multiple electrical signals in the welding circuit where the weld seam is located during the welding process at the welding station. A controller, connected to the acquisition circuit, is used to divide the multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of the multiple electrical signals, extract time-frequency features of target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld, and perform quality inspection on the weld based on the multiple time-frequency features to obtain an inspection result. The signal values ​​of the multiple target signals in the target signal segments are within a preset range, and the duration for which the signal values ​​of the multiple target signals are within the preset range reaches a preset time. The inspection result includes the quality level of the weld.

14. The circuit according to claim 13, characterized in that, The acquisition circuit includes: A sensor, connected to the weld, is used to collect multiple analog signals in the welding circuit; A filter, located between the sensor and the data acquisition card, is used to filter the multiple analog signals to obtain multiple processed analog signals; A data acquisition card, located between the filter and the controller, is used to convert the plurality of processed analog signals into the plurality of electrical signals.

15. The circuit according to claim 14, characterized in that the sensor include: A voltage sensor, connected in parallel with the weld, is used to collect multiple voltage signals; A current sensor, connected in series with the weld, is used to collect multiple current signals; The plurality of analog signals include the plurality of voltage signals and the plurality of current signals.

16. A weld inspection system, characterized in that, include: The signal acquisition module is used to acquire multiple electrical signals in the welding circuit where the weld is located during the welding process at the welding station. The feature transformation module is used to divide the multiple electrical signals into sub-signal segments of different time periods based on the signal values ​​of the multiple electrical signals, and to extract time-frequency features of the target signal segments in the sub-signal segments of different time periods to obtain multiple time-frequency features of the weld. The signal values ​​of the multiple target signals in the target signal segments are within a preset range, and the duration for which the signal values ​​of the multiple target signals are within the preset range reaches a preset time. The quality assessment module is used to perform quality inspection on the weld based on the multiple time-frequency features and obtain inspection results, wherein the inspection results include the quality grade of the weld.

17. A vehicle, characterized in that, include: The vehicle body, wherein the welds on the vehicle body are detected by the detection method of any one of claims 1-12, or by the detection circuit of any one of claims 13-15, or by the detection system of claim 16.

18. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 12.

Images