Power generation welding machine welding quality detection method and system based on dynamic resistance curve

Abstract

The present application relates to the technical field of resistance measurement, in particular to a power generator welding machine welding quality detection method and system based on dynamic resistance curve, which synchronously collects instantaneous terminal voltage, instantaneous loop current and generator rotor phase pulse sequence; additional angle label is added and equal-angle resampling is carried out, time domain sequence is converted into fixed angle interval sequence; dynamic resistance curve is reconstructed by point-by-point division, reference dynamic resistance curve is generated by sliding median filtering and zero phase shift filtering; difference positioning feature inflection point is carried out and the curve is divided into contact resistance decay section, fusion core resistance rising section and fusion core resistance falling section; average slope and peak resistance are extracted from the rising section, falling rate and integral area are extracted from the falling section to form a multi-dimensional feature vector; welding quality index is generated by weighted fusion and normalization, and compared with standard interval to determine qualified, underwelding or overburning. The influence of speed disturbance on resistance detection is eliminated, welding is determined immediately, and the detection accuracy and on-site practicability are improved.

Description

Technical Field

[0001] This invention relates to the field of resistance measurement technology, and in particular to a method and system for detecting the welding quality of a power generation welding machine based on a dynamic resistance curve. Background Technology

[0002] In field pipeline laying, emergency steel structure repair, and construction work in areas without mains power, generator-welded machines (i.e., integrated devices that use an internal combustion engine to drive a generator as the welding power source) are widely used due to their portability and independence. Currently, the mainstream methods for online inspection of the welding quality of generator-welded machines include post-weld destructive sampling inspection, offline inspection based on ultrasonic or radiographic testing, and routine statistical monitoring based on the effective values ​​of welding voltage and welding current.

[0003] However, existing detection technologies have significant limitations. Destructive and offline flaw detection methods cannot achieve "immediate assessment upon welding," leading to severe delays in the rework of substandard welds and significantly increasing construction costs. Conventional voltage and current RMS monitoring only focuses on the average level of macroscopic electrical parameters, completely missing the rich transient information contained in the droplet transition, dynamic shrinkage of the molten pool, and changes in resistive heating effects. Especially for generator welding machines, the inherent speed fluctuations and excitation regulation response lag of the internal combustion engine result in a large number of low-frequency disturbance signals mixed in the welding circuit. Traditional dynamic resistance detection methods are prone to characteristic distortion because they cannot effectively distinguish between "the physical resistance change of the weld nugget itself" and "false resistance fluctuations caused by generator output fluctuations," leading to serious misjudgments and missed detections. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for detecting the welding quality of a generator welding machine based on a dynamic resistance curve, which significantly improves the accuracy and real-time performance of the detection of welding quality in generator welding machines.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for inspecting the welding quality of a power generation welding machine based on a dynamic resistance curve, comprising the following steps:

[0006] During welding, the instantaneous terminal voltage signals at both ends of the workpiece, the instantaneous circuit current signal flowing through the welding circuit, and the phase pulse sequence of the generator rotor of the generator welding machine are synchronously collected at a fixed time sampling frequency.

[0007] Hardware filtering and analog-to-digital conversion are performed on the acquired instantaneous terminal voltage signal and instantaneous loop current signal. Angle labels are added to the converted voltage and current data based on the phase pulse sequence to establish the correspondence between each sampling data point and the spatial angle of the generator rotor.

[0008] Based on the angle label, the sampling sequence with a fixed time interval is converted into a resampled terminal voltage sequence and a resampled loop current sequence with a fixed angle interval.

[0009] A point-by-point division operation is performed on the resampled terminal voltage sequence and the resampled loop current sequence to obtain the original dynamic resistance curve that varies with the angle. The original dynamic resistance curve is then subjected to sliding median filtering and zero phase shift low-pass filtering to generate a reference dynamic resistance curve.

[0010] Based on the differential characteristics of the reference dynamic resistance curve, three key feature inflection points are automatically located, and the reference dynamic resistance curve is divided into a contact resistance attenuation segment, a molten core resistance rising segment, and a molten core resistance falling segment using the three key feature inflection points as dividing boundaries.

[0011] The average slope value and peak resistance value are extracted from the rising segment of the molten core resistance, and the falling rate value and integral area value are extracted from the falling segment of the molten core resistance. The four feature values ​​together constitute a multidimensional original feature vector.

[0012] A preset welding quality quantification model is invoked to perform weighted fusion and normalization operations on the multidimensional original feature vectors to generate a real-time welding quality index.

[0013] The real-time welding quality index is compared with the standard quality index range to determine whether the current weld quality is qualified, under-welded, or over-burned, and the corresponding determination signal is output.

[0014] During welding energization, the instantaneous terminal voltage signals at both ends of the workpiece, the instantaneous circuit current signal flowing through the welding circuit, and the phase pulse sequence of the generator rotor of the welding machine are synchronously acquired at a fixed sampling frequency, including:

[0015] The voltage sampling channel, current sampling channel, and rotor phase encoding acquisition channel are driven to operate synchronously by the master clock signal generated by the same high-stability clock source, ensuring that the sampling time deviation between channels is controlled within 1 microsecond; the phase pulse sequence is generated by an incremental photoelectric encoder coaxially connected to the generator rotor, and each phase pulse is triggered by a rising edge and the corresponding global timer count value is latched as a timestamp.

[0016] The hardware filtering uses a second-order Butterworth active low-pass filter with a cutoff frequency of 3 kHz to filter out high-frequency switching noise; the analog-to-digital conversion uses a 16-bit successive approximation analog-to-digital converter to perform synchronous digitization at a sampling rate of 200 kilos per second per channel.

[0017] The process of adding angle labels is as follows: taking the fixed angle interval between two adjacent rising edges of phase pulses as an angle interval unit, the timestamps of each voltage data and current data in the angle interval unit are converted into the corresponding angle positions in the angle interval unit according to the linear proportional distribution principle, thereby establishing a one-to-one correspondence between each sampled data point and the rotor spatial angle.

[0018] Specifically, based on the angle label, converting the sampling sequence with a fixed time interval into a resampled terminal voltage sequence and a resampled loop current sequence with a fixed angle interval includes:

[0019] Using the base interval determined by the number of pulses per revolution of the encoder as the original angle interval unit, several equally spaced target angle grid points are generated within each original angle interval unit. The original sampled data falling within the original angle interval unit is interpolated and reconstructed using a piecewise cubic conformal interpolation algorithm. The reconstructed terminal voltage value and loop current value are obtained at each target angle grid point. Finally, the end-to-end are spliced ​​together to form a resampled terminal voltage sequence and a resampled loop current sequence with a fixed angle interval.

[0020] Specifically, the point-by-point division operation is performed one by one according to the same angle index position, and the set current points are invalidated and replaced using the current lower limit threshold; the sliding median filtering adopts a filter window with a preset angle width, sorts the resistance values ​​in the window and takes the median as the output value of the window center point, which is used to filter out the spikes and glitches caused by droplet short-circuit transition; the zero-phase-shift low-pass filtering adopts a cascaded method of forward filtering and reverse filtering, which ensures that the position of the key feature inflection point on the angle axis does not shift while smoothing the residual ripple.

[0021] The specific methods for locating the three key feature inflection points are as follows: The first feature inflection point is located at the peak point of solid-state resistance when the electrode contacts the workpiece at the beginning of the welding process, determined by finding the zero-crossing point where the first derivative changes from positive to negative within a preset first search window; the second feature inflection point is located at the turning point of the resistance rise rate after the rapid expansion of the weld nugget ends, determined by monitoring the position where the first derivative falls back from its maximum value to a preset proportional threshold within a second search window after the first feature inflection point; the third feature inflection point is located at the starting point of resistance decay as the weld nugget shrinks towards the liquid bridge after its growth is complete, determined by finding the zero-crossing point where the second derivative changes from negative to positive within a third search window after the second feature inflection point.

[0022] The average slope of the rising section of the molten core resistance is obtained by performing a univariate linear regression trend analysis on all discrete data points within the section, and is used to characterize the rate of thermal accumulation of the molten core resistance. The peak resistance value is obtained by taking the maximum value of all resistance values ​​within the section, and is used to characterize the maximum resistance extreme value when the volume of the molten core expands to its limit. The falling rate of the falling section of the molten core resistance is obtained by calculating the resistance difference between the starting point and the ending point of the section and dividing it by the angular width it spans, and is used to characterize the speed at which the molten core contracts towards the liquid bridge. The integral area value is obtained by summing the areas of the trapezoids formed by adjacent data points within the section, and is used to characterize the overall heat dissipation characteristics during the molten core contraction stage.

[0023] The construction process of the welding quality quantification model includes: selecting standard specimens for welding under standard process parameters, extracting multidimensional original feature vectors of qualified weld points as positive sample sets, determining the weight coefficients corresponding to the four features through multiple regression analysis to form a multidimensional feature weight matrix, and calculating the arithmetic mean of the fusion values ​​of the positive sample set as the central benchmark value and the standard deviation as the benchmark dispersion; the weighted fusion is to multiply the four feature values ​​of the current weld point by the corresponding weight coefficients and then sum them to obtain the original fusion total value; the normalization operation is to subtract the central benchmark value from the original fusion total value, divide it by the benchmark dispersion, and then perform nonlinear amplitude limiting mapping to obtain a dimensionless integer between zero and one hundred as the real-time welding quality index.

[0024] The standard quality index range is determined based on the index distribution statistics of the qualified positive sample set combined with the engineering redundancy margin, and has a clear lower limit boundary value and an upper limit boundary value. When the real-time welding quality index falls within the standard quality index range, the current weld point is determined to be qualified and a qualified confirmation signal is output. When the real-time welding quality index is lower than the lower limit boundary value, the current weld point is determined to be under-welded and an under-welded alarm signal is output. When the real-time welding quality index is higher than the upper limit boundary value, the current weld point is determined to be over-burned and an over-burned alarm signal is output. The under-welded alarm signal and the over-burned alarm signal are synchronously output to the protection control circuit of the generator welding machine to trigger the welding main circuit to be cut off and the welding parameters of the next weld point to be automatically corrected.

[0025] Secondly, the present invention provides a welding quality inspection system for a power generation welding machine based on a dynamic resistance curve, used to implement a welding quality inspection method for a power generation welding machine based on a dynamic resistance curve as provided in the first aspect, comprising:

[0026] The voltage sampling module and the current sampling module are connected in parallel and in series to the welding main circuit of the generator-weld machine, respectively, to synchronously collect the instantaneous terminal voltage signal at both ends of the workpiece and the instantaneous circuit current signal flowing through the welding circuit during welding power-on.

[0027] The rotor phase encoding unit is coaxially connected to the generator rotor of the generator welding machine and is used to output a phase pulse sequence corresponding to the rotor rotation angle;

[0028] The signal preprocessing unit, connected to the voltage sampling module, the current sampling module and the rotor phase encoding unit, is used to perform hardware filtering, analog-to-digital conversion and timestamp calibration on the acquired instantaneous terminal voltage signal and instantaneous loop current signal;

[0029] An equal-angle resampling unit, connected to the signal preprocessing unit, is used to convert a sampling sequence with a fixed time interval into a resampling terminal voltage sequence and a resampling loop current sequence with a fixed angle interval based on the phase pulse sequence.

[0030] The dynamic resistance curve reconstruction unit, connected to the equal-angle resampling unit, is used to perform point-by-point division and composite filtering on the resampling terminal voltage sequence and the resampling loop current sequence to generate a reference dynamic resistance curve.

[0031] The feature recognition and segmentation unit, connected to the dynamic resistance curve reconstruction unit, is used to automatically locate three key feature inflection points and segment the reference dynamic resistance curve into a contact resistance attenuation segment, a molten core resistance rising segment, and a molten core resistance falling segment.

[0032] The feature extraction unit, connected to the feature recognition and segmentation unit, is used to extract the average slope value and peak resistance value from the rising segment of the molten core resistor, and the falling rate value and integral area value from the falling segment of the molten core resistor, to generate a multi-dimensional original feature vector;

[0033] The quality assessment unit, built into the generator welding machine controller and connected to the feature extraction unit, is used to call the welding quality quantification model to perform weighted fusion and normalization on the multi-dimensional original feature vector to generate a real-time welding quality index, and compare the real-time welding quality index with the standard quality index range to determine the quality, and output the corresponding quality judgment signal to the display terminal and protection control loop.

[0034] This invention discloses a welding quality detection method and system for generator welding machines based on dynamic resistance curves. During welding energization, the system simultaneously acquires instantaneous terminal voltage, instantaneous loop current, and generator rotor phase pulse sequences. After adding angle tags to the acquired signals, it performs equal-angle resampling, converting the fixed-time interval sampling sequence into a fixed-angle interval sequence. The original dynamic resistance curve is reconstructed by point-by-point division, and a composite filter is performed to generate a reference dynamic resistance curve. Three key feature inflection points are automatically located, dividing the curve into a contact resistance attenuation segment, a weld nugget resistance rise segment, and a weld nugget resistance fall segment. The average slope value and peak resistance value are extracted from the rise segment, and the fall rate value and integral area value are extracted from the fall segment to construct a multi-dimensional original feature vector. A welding quality quantification model is called to perform weighted fusion and normalization to generate a real-time welding quality index. The real-time welding quality index is compared with a standard quality index range to determine whether the weld is qualified, under-welded, or over-burned, and a corresponding signal is output. This invention eliminates speed disturbances by introducing rotor phase equal-angle resampling and constructs a multi-dimensional fused quality index to achieve immediate judgment after welding, significantly improving the accuracy and real-time performance of generator welding machine welding quality detection. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0036] Figure 1 This is a schematic diagram of the steps of a welding quality detection method for a generator welding machine based on dynamic resistance curves according to the first embodiment of the present invention.

[0037] Figure 2 This is a schematic flowchart of a welding quality inspection method for a generator welding machine based on dynamic resistance curves provided by the present invention.

[0038] Figure 3 This is a schematic diagram of the structural principle of a power generation welding machine welding quality detection system based on dynamic resistance curve according to the second embodiment of the present invention. Detailed Implementation

[0039] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0040] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0041] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0042] The first embodiment of this application is as follows:

[0043] Please see Figures 1-3 This invention provides a method for inspecting the welding quality of a power generation welding machine based on a dynamic resistance curve, comprising the following steps:

[0044] S1. During welding energization, the instantaneous terminal voltage signals at both ends of the workpiece, the instantaneous circuit current signals flowing through the welding circuit, and the phase pulse sequence of the generator rotor of the generator welding machine are synchronously collected at a fixed time sampling frequency.

[0045] Specifically, it relies on a high-precision synchronous data acquisition hardware architecture, which consists of a system-level high-stability clock source, voltage sampling probe, current sampling probe, rotor phase encoder, and multi-channel synchronous analog-to-digital converter front end.

[0046] After the system is powered on and initialized, the high-stability clock source built into the generator-welder controller immediately starts, generating a 10 MHz master clock signal. This master clock signal is fanned out by the clock drive circuit into three synchronous trigger clocks with the same frequency and phase, which are respectively distributed to the voltage sampling channel, the current sampling channel, and the rotor phase encoding acquisition channel. This ensures that the sampling actions of these three channels are strictly controlled to the same time base, keeping the inherent sampling time deviation between channels within 1 microsecond.

[0047] During welding, the voltage sampling probe employs a high-input-impedance differential amplification structure. Its two input terminals are connected to the working and induction electrodes at both ends of the weldment via shielded wires, directly picking up the instantaneous potential difference between the two electrodes to form an instantaneous terminal voltage analog signal. Simultaneously, the current sampling probe uses a through-hole Hall effect current sensor. The main cable of the welding circuit under test is passed directly through the sensor's central aperture, utilizing the Hall effect to non-contactly induce an analog voltage output proportional to the instantaneous circuit current, forming an instantaneous circuit current analog signal. Both analog signals are transmitted to the signal conditioning board via independently shielded twisted-pair cables to suppress external electromagnetic interference.

[0048] Simultaneously, the rotor phase encoder, rigidly coaxially connected to the main shaft of the generator welding machine's internal combustion engine, begins operation. This encoder is an incremental photoelectric encoder, with a pulse count set to 720 per revolution, meaning it outputs 720 phase pulses per revolution of the generator rotor (360° electrical angle). The phase pulse sequence output by the encoder is sent to the acquisition front end as a digital interrupt signal after passing through a high-speed opto-isolator. At the moment welding begins, the controller sends a start acquisition command to the acquisition front end, which then activates a global uninterrupted sampling timer based on the master clock. Subsequently, for each data point acquired by the voltage and current sampling channels at each sampling moment, the analog-to-digital converter automatically reads the current count value from the global sampling timer at the moment of conversion and stores it as a unique timestamp for that data point in the buffer. For each rising edge of a pulse in the phase pulse sequence, the system also responds with an interrupt, immediately reading the current count value of the global sampling timer at the moment the rising edge is triggered and latching it as the timestamp for that pulse.

[0049] Through the triple mechanism of driving with the same clock source, synchronous triggering acquisition, and global timestamp calibration, an initial correspondence based on absolute time is established among the instantaneous terminal voltage signal, instantaneous loop current signal, and phase pulse sequence, laying a solid original data foundation for subsequent accurate mapping from the time domain to the angle domain.

[0050] S2. Perform hardware filtering and analog-to-digital conversion on the acquired instantaneous terminal voltage signal and instantaneous loop current signal, and add angle labels to the converted voltage data and current data based on the phase pulse sequence to establish the correspondence between each sampling data point and the generator rotor spatial angle.

[0051] Specifically, the first step involves front-end conditioning and hardware filtering of the analog signals. The two analog signals from the voltage and current sampling probes enter the signal conditioning board and first pass through an impedance matching and amplitude conditioning submodule. For instantaneous terminal voltage signals, this submodule uses a precision resistor divider network to attenuate their amplitude to a standard ±10V input voltage range suitable for the analog-to-digital converter. For instantaneous loop current signals, this submodule uses a current-to-voltage conversion circuit composed of a precision operational amplifier to amplify the weak voltage signal output from the Hall sensor to a standard ±10V range. After amplitude conditioning, the two analog signals are then fed into a second-order Butterworth active low-pass filter for hardware filtering. The cutoff frequency of this filter is set to 3 kHz. Its function is twofold: firstly, it allows signal components with frequencies below 3 kHz that reflect the resistive heating effect and changes in the molten droplet transition body during the formation of the weld nugget to pass through without attenuation; secondly, it provides an attenuation slope of approximately 40 dB per octave for high-frequency components with frequencies above 3 kHz—such as the high-frequency chopping noise generated by the operation of power switching devices in the inverter circuit of the generator welding machine, the high-frequency radiated interference generated during the combustion of the welding arc, and the high-frequency spatial interference coupled during cable transmission—thereby significantly reducing the adverse effects of these high-frequency noises on the accuracy of subsequent dynamic resistance calculations.

[0052] Next, analog-to-digital conversion and digital timestamp calibration are performed. The two analog signals, after hardware filtering, are respectively input to two independent input channels of a 16-bit successive approximation analog-to-digital converter (ADC). Under the control of the synchronous trigger clock described in the first step, the ADC synchronously digitizes the instantaneous terminal voltage analog signal and the instantaneous loop current analog signal at a sampling rate of 200 kilos per second per channel (i.e., one sampling conversion every 5 microseconds). After each conversion, the ADC outputs the corresponding digital instantaneous voltage and current values. Simultaneously with writing the data to the buffer, the system reads the current count value from the global sampling timer as the joint timestamp of the voltage and current data set, and binds this timestamp with the digital voltage and current values ​​to form a complete data record unit. Through this process, each voltage and current record in the buffer precisely corresponds to an absolute time coordinate on the order of microseconds.

[0053] Finally, the angle mapping relationship is calculated and established based on the phase pulse timestamps. After processing the aforementioned digitized voltage and current data, the system retrieves all recorded phase pulse sequence timestamp data from the buffer. Since the encoder has 720 pulses per revolution, the mechanical angle difference between the rising edges of two adjacent phase pulses is fixed at 0.5° (i.e., 360° divided by 720). The system sequentially retrieves the timestamps of the rising edges of two adjacent phase pulses according to their chronological order and defines the time interval between these two pulses as an "angle interval unit". Within this angle interval unit, the system searches all voltage and current data recording units cached in the first step and extracts the timestamp from each record one by one. Based on the linear proportional allocation principle, the system determines the relative position of the timestamp within the time interval of two adjacent pulses: if the timestamp is exactly equal to the timestamp of the previous pulse, it is assigned a corresponding integer multiple angle value; if the timestamp is between two pulse timestamps, it is increased by the corresponding angle increment based on the angle of the previous pulse according to the time ratio. After the above-mentioned retrieval and assignment calculation, each voltage sampling point and current sampling point in the buffer obtains a corresponding rotor space electrical angle label. Thus, the instantaneous terminal voltage signal and instantaneous loop current signal, which originally only had time coordinates, were successfully supplemented with angular coordinates that corresponded one-to-one with the actual spatial position of the generator rotor, establishing a precise "time-angle" dual-domain mapping relationship. This mapping relationship ensures that even when the internal combustion engine speed fluctuates drastically, causing significant changes in the interval between adjacent pulses, the system can still accurately determine which spatial position of the rotor each sampling point corresponds to, thus providing an accurate index for the equal-angle resampling in the third step.

[0054] S3. Based on the angle label, convert the sampling sequence with a fixed time interval into a resampled terminal voltage sequence and a resampled loop current sequence with a fixed angle interval.

[0055] Specifically, firstly, the target angle grid is determined and angle interval units are divided. Based on the number of pulses per revolution of the rotor phase encoder in the second step, the system sets the target angle step size for equal-angle resampling. Although the original encoder pulse interval is 0.5° mechanical angle, to improve the resolution of the subsequent dynamic resistance curve and avoid the step effect caused by excessively coarse pulse intervals, the system sets the target angle step size to 0.1°. This means the system expects to obtain 3600 equally spaced sampling data points within a complete 360° mechanical rotation cycle. Subsequently, the system marks each angle interval defined by adjacent phase pulses within the entire acquisition cycle as a basic processing unit. The starting boundary of each processing unit is the angle value corresponding to the rising edge of the previous phase pulse, and the ending boundary is the angle value corresponding to the rising edge of the next phase pulse. The total span of this unit is a fixed 0.5° mechanical angle, but due to speed fluctuations, the time span corresponding to this unit changes in real time.

[0056] Secondly, for each angle interval unit, the original sampling points and angle labels within it are extracted. The system traverses all original voltage and current data records with angle labels in the buffer. For each 0.5° span angle interval unit, the system retrieves all original sampling points whose angle labels fall between the start and end angles of that unit. Since each sampling point has been assigned a precise angle label in the second step, these original sampling points are not uniformly distributed on the angle axis—when the internal combustion engine accelerates, the distribution of sampling points on the angle axis is sparser (short time intervals, large angle spans); when the internal combustion engine decelerates, the distribution is denser (long time intervals, small angle spans). The system sorts these original sampling points according to their angle label values ​​from smallest to largest, forming a "original angle - signal value" correspondence sequence within that unit. This sequence contains both the instantaneous terminal voltage original value and the instantaneous loop current original value.

[0057] Next, a conformal interpolation algorithm is used to reconstruct the signal value at the target angle grid points. The system generates five equally spaced target angle grid points (i.e., starting angle, starting angle +0.1°, starting angle +0.2°, starting angle +0.3°, starting angle +0.4°, and ending angle) within each 0.5° original angle interval unit, using a pre-set uniform interval of 0.1°. The ending angle is reused as the starting point of the next unit. For each target angle grid point, the system performs piecewise cubic conformal interpolation using the "angle-signal value" correspondence between several adjacent original sampling points. The core constraint of this interpolation method is that the interpolated curve not only passes through all original sampling points but also maintains the continuity of the first derivative at each original sampling point, while strictly ensuring that no additional spurious extrema are generated within the interval between two adjacent original sampling points. Compared to ordinary linear interpolation, conformal interpolation can more realistically reproduce the physical trend of resistance change with angle during welding; compared to conventional cubic spline interpolation, it effectively avoids overshoot or undershoot caused by drastic data fluctuations. The system independently performs the above conformal interpolation process on the original instantaneous terminal voltage and the original instantaneous loop current, respectively, and obtains the reconstructed instantaneous terminal voltage value and the reconstructed instantaneous loop current value at each target angle grid point.

[0058] Finally, the standard sequence after equal-angle resampling is output. After the above processing of each angle interval unit, the system splices the target angle grid points in all units end to end in ascending order of angle value to form a complete resampled terminal voltage sequence and a resampled loop current sequence. These two sequences share common characteristics: each data point in the sequence strictly corresponds to a fixed mechanical angle interval of 0.1°, and the length of the entire sequence is a fixed value within any welding energizing cycle (for example, if the generator rotates twice, the sequence length is always 7200 points). Thus, the stretching and compression effects caused by the transient changes in internal combustion engine speed in the original acquisition sequence are completely eliminated. Even if the generator-welder experiences sudden acceleration or deceleration during welding, the resampled voltage and current waveforms maintain a stable rhythm in the angle domain, fundamentally ensuring the authenticity and comparability of the subsequent dynamic resistance curve shape.

[0059] S4. Perform point-by-point division on the resampled terminal voltage sequence and the resampled loop current sequence to obtain the original dynamic resistance curve that varies with the angle, and perform sliding median filtering and zero phase shift low-pass filtering on the original dynamic resistance curve to generate a reference dynamic resistance curve.

[0060] Specifically, firstly, the original dynamic resistance curve is reconstructed by dividing point by point. The system extracts the resampled terminal voltage sequence and resampled loop current sequence obtained in the third step, one-to-one according to the same angular index position. For each angular index point, the system performs a numerical calculation using Ohm's law, that is, dividing the instantaneous terminal voltage value at that point by the instantaneous loop current value at that point, and the quotient is the transient resistance value at that angular position. Since the resampled terminal voltage and resampled loop current are strictly synchronized and correspond one-to-one in the angular domain, a raw dynamic resistance curve with the angle as the horizontal axis and the transient resistance value as the vertical axis is directly obtained after point-by-point division. It should be noted that, since the current in the welding circuit may reach a very small value close to zero at the moment of short-circuit transition, in order to avoid numerical overflow, the system sets a current lower limit threshold (e.g., 5 amps) close to zero before the division operation. When the absolute value of the measured current is lower than this lower limit threshold, the system marks the resistance value of that point (the set current point) as an invalid point, and replaces it in subsequent calculations by interpolation of adjacent valid points. Through the above point-by-point calculations, the system fully preserves the instantaneous change details of the fusion core resistance at each angular position.

[0061] Secondly, a sliding median filtering algorithm is used to filter out spikes and glitches caused by the short-circuit transition of the molten droplet. Although the baseline drift caused by the rotation speed disturbance has been eliminated from the original dynamic resistance curve, the molten droplet will cause the welding circuit to be approximately in a direct short-circuit state at the moment of short-circuit transition, resulting in a very narrow steep drop peak in the calculated resistance value. In order to filter out these non-physical isolated peaks without damaging the overall shape of the curve, the system first performs sliding median filtering. The system sets the angle width of the filtering window to 0.7° (corresponding to 7 equal angle sampling points, since the target step size is 0.1°). The filtering window starts from the starting point of the original dynamic resistance curve and slides to the right one angle point by one in a step size of 0.1°. At each window position, the system reads the transient resistance values ​​corresponding to the 7 consecutive angle points covered by the window, sorts these 7 values ​​in ascending order, and takes the 4th value in the middle position after sorting as the filtered output value at the center point of the window. The core advantage of sliding median filtering lies in the fact that the spikes generated by droplet short circuits typically appear as outliers with maximum or minimum values. After sorting, these outliers are pushed to the ends of the sequence, while the values ​​in the middle of the sequence precisely represent the mainstream resistance values ​​within that local interval that are unaffected by spikes. This filtering process can cleanly and efficiently remove short-circuit spikes completely, while preserving the steepness of the rising and falling edges of the resistance curve, unlike linear mean filtering which blurs edge features.

[0062] Finally, a zero-phase-shift low-pass filter is performed to eliminate residual minor mechanical vibration noise. After the sliding median filter, the large isolated peaks in the curve have been effectively eliminated, but the inherent mechanical torsional vibration and bearing movement during generator rotor rotation may still superimpose small-amplitude periodic ripple noise on the resistance curve. To further improve the smoothness of the curve while ensuring that the filtering process does not introduce any phase delay—because phase delay will directly cause the angle positioning of the three key feature inflection points in step five to shift, ultimately affecting the accuracy of quality judgment—the system adds a zero-phase-shift low-pass filter after the sliding median filter. The specific implementation of this step is as follows: the system first inputs the dynamic resistance sequence after median filtering into a digital low-pass filter (whose equivalent cutoff frequency corresponds to a high-frequency ripple that changes more than 5 times per angle period) in the forward direction of the angle to obtain the first round of filtering results; then, the first round of filtering results is input into the same digital low-pass filter again in the reverse direction of the angle (i.e., from the end of the curve to the beginning) to obtain the second round of filtering results. The cascaded execution of forward and reverse filtering ensures that the phase lead introduced by the forward filter is precisely offset by the phase lag introduced by the reverse filter, resulting in a final output curve with no net phase shift on the angle axis. After this zero-phase-shift processing, the subtle periodic ripples in the curve are fully smoothed, while the positions of the three key features—the extreme points of the rising slope, the extreme points of the peak resistance, and the inflection points of the falling curvature—are strictly maintained on their original coordinates in the angle domain without any translation.

[0063] After processing using the composite filtering strategy consisting of coarse-grained peak removal (sliding median filtering) and fine-grained ripple smoothing (zero-phase-shift low-pass filtering), the system ultimately generates a benchmark dynamic resistance curve with an extremely high signal-to-noise ratio, a smooth and natural curve shape, and zero deviation in the position of key feature inflection points. This curve not only realistically reflects the resistive thermal evolution law of the weld nugget from formation to shrinkage, but also provides accurate and reliable data support for the fifth step of automatic feature inflection point identification based on the zero-crossing point of the second derivative.

[0064] S5. Based on the differential characteristics of the reference dynamic resistance curve, automatically locate three key feature inflection points, and use the three key feature inflection points as dividing boundaries to divide the reference dynamic resistance curve into a contact resistance attenuation segment, a molten core resistance rising segment, and a molten core resistance falling segment.

[0065] Specifically, firstly, numerical difference operations are performed on the reference dynamic resistance curve to obtain the rate of change information. The system sends the reference dynamic resistance curve, with the angle axis as the independent variable and the resistance value as the dependent variable, to the feature recognition and segmentation submodule. This submodule performs a first-order numerical difference operation on the curve point by point, that is, calculates the instantaneous rate of change of the resistance value relative to the angle change at each angular position on the curve, and obtains a first-order derivative discrete sequence. The sign and magnitude of this sequence intuitively reflect whether the resistance is in an upward trend, a downward trend, or a flattening trend at the current angle. At the same time, the system performs a difference operation again on the first-order derivative discrete sequence to obtain a second-order derivative discrete sequence. This sequence reveals the inherent variation law of the rate of change of the resistance curve, that is, whether the upward trend of the resistance is accelerating or decelerating, and whether the downward trend is intensifying or slowing down.

[0066] Secondly, an adaptive search window is set to narrow the traversal range of feature inflection points. The system does not blindly search across the entire angle domain, but rather, based on the physical temporal laws of the weld nugget formation process, presets three non-overlapping adaptive search windows on the angle axis. Specifically, the system considers the total angle span corresponding to the entire welding energizing cycle as a 100% baseline. The first search window is set within the first 5% to 15% of the total angle span, corresponding to the initial contact stage after welding starts, where the electrode interacts with the oxide film and micro-roughness peaks on the workpiece surface. The second search window is set within the 20% to 65% of the total angle span, corresponding to the main stage of nugget volume expansion and resistance heat accumulation. The third search window is set within the 65% to 90% of the total angle span, corresponding to the later stage where the nugget reaches its maximum volume and begins to shrink into the molten pool. The system uses adaptive search windows because, although the nugget formation rate varies under different weldment specifications and welding current conditions, the order and relative proportion of the three metallurgical stages on the time or angle axis are statistically stable. If the system fails to detect a valid feature inflection point within the preset window, the window boundary will gradually expand outward in steps of 0.5% of the total angular span until the inflection point is successfully captured, thus ensuring adaptability to different welding process parameters.

[0067] Next, within their respective search windows, the precise angular positions of the three key feature inflection points are located based on the differential features. The system executes different discrimination logic in each of the three search windows:

[0068] To locate the first characteristic inflection point, the system first scans the first derivative sequence within the first search window, searching for the zero-crossing point where the first derivative changes from positive to negative, and uses the corresponding angular position as a candidate point. This candidate point represents the critical position where the resistance value changes from an initial state of continuous increase to a state of decrease, which corresponds precisely to the maximum solid-state contact resistance before the oxide film and micro-roughness peaks on the electrode and workpiece surface are crushed and broken down, i.e., the peak point of solid-state resistance. The system locks the position of this candidate point as the precise angular coordinate of the first characteristic inflection point.

[0069] To locate the second feature inflection point, the system uses the angular position of the first feature inflection point as a starting point and continuously tracks the changing trend of the first derivative within a second search window. During the rapid expansion phase of the molten nucleus, the resistance increases sharply with rising temperature, and the first derivative maintains a large positive value. As the molten nucleus expansion gradually approaches its limit, the rate of increase in resistance begins to slow down, and the value of the first derivative continuously declines from its peak. The system monitors the first derivative sequence, and when its value drops from the maximum value within the window to below a preset percentage threshold (e.g., 25%) of that maximum value, the system locks the angular position corresponding to this decline point as the second feature inflection point. This inflection point accurately reflects the turning point when the rapid expansion of the molten nucleus ends and the rate of increase in resistance declines.

[0070] To locate the third characteristic inflection point, the system focuses on the sign change of the second derivative sequence within the third search window following the second characteristic inflection point. In the initial stage of the molten core resistance decrease phase, the resistance value continuously decreases at an increasingly rapid rate, at which point the second derivative is negative. As the molten core further contracts towards the liquid bridge and stabilizes, the resistance decrease rate gradually slows down, and the second derivative gradually recovers from its negative value. The system searches for the zero-crossing point where the second derivative changes from negative to positive, and locks the corresponding angular position as the third characteristic inflection point. This inflection point precisely corresponds to the starting position where the molten core growth is basically complete and the resistance decay trend begins to narrow.

[0071] Finally, the reference dynamic resistance curve is divided into three characteristic segments based on three characteristic inflection points. After determining the precise angular coordinates of the first, second, and third characteristic inflection points, the system uses these three inflection points as dividing boundaries to perform segmentation operations on the entire reference dynamic resistance curve. Specifically, the segmentation is as follows: the curve segment from the initial angle of the curve (the moment welding energization begins) to the first characteristic inflection point is designated as the contact resistance decay segment, reflecting the physical process of the surface contact resistance between the electrode and the workpiece gradually decreasing from its peak value; the curve segment from the first to the second characteristic inflection point is designated as the weld nugget resistance rise segment, reflecting the main welding stage where the weld nugget expands in volume under resistive heating, and the resistance continuously increases with temperature; the curve segment from the second to the third characteristic inflection point is designated as the weld nugget resistance fall segment, reflecting the final stage where the weld nugget volume reaches its maximum and begins to shrink towards the liquid bridge, the conductive cross-section increases, and the resistance value gradually decreases. The three sections mentioned above are sequential and do not overlap, completely covering the dynamic resistance change trajectory of the entire welding energization cycle.

[0072] S6. Extract the average slope value and peak resistance value from the rising segment of the molten core resistor, and extract the falling rate value and integral area value from the falling segment of the molten core resistor. The four feature values ​​together constitute a multidimensional original feature vector.

[0073] Specifically, firstly, the system extracts the average slope and peak resistance values ​​from the rising segment of the molten core resistance. The system sends the rising segment of the molten core resistance (i.e., all discrete data points between the first and second feature inflection points) extracted in step five to the feature extraction submodule. For the extraction of the average slope, the system performs a trend analysis in the sense of univariate linear regression on all discrete data points within this segment: it calculates the angular offset of each data point relative to the first feature inflection point and its corresponding resistance increment, sums the covariance of the angular offsets and resistance increments of all data points, and divides the sum by the sum of the squares of the angular offsets. The quotient is the average slope value of this rising segment. This average slope value quantifies the rate of volume expansion of the molten core under resistive heating. A high slope value indicates rapid resistive heat accumulation and a fast temperature rise rate in the molten core, while a low slope value indicates insufficient heat input or excessive heat dissipation. For the extraction of the peak resistance value, the system directly iterates through the resistance values ​​of all data points within the rising segment of the molten core resistance, finds the maximum value, and the resistance value corresponding to this maximum value is the peak resistance value. Since this peak value occurs precisely near the second characteristic inflection point, it represents the maximum resistance extreme value that the molten nugget can reach when its volume expands to the limit state, directly reflecting the highest resistance thermal state experienced by the molten nugget during the welding process.

[0074] Secondly, the descent rate and integral area values ​​are extracted from the decreasing resistance segment of the molten core. The system sends the decreasing resistance segment of the molten core extracted in step five (i.e., all discrete data points between the second and third feature inflection points) into the same feature extraction submodule. For the extraction of the descent rate value, the system calculates the resistance difference between the starting point (i.e., the resistance value at the second feature inflection point) and the ending point (i.e., the resistance value at the third feature inflection point) of the decreasing segment, and divides this difference by the total angular width spanned by the decreasing segment. The absolute value of the quotient is the average descent rate value of the decreasing segment. This descent rate value reflects the rate of contraction of the molten core as it transforms from a solid bridge to a liquid bridge: an excessively fast descent rate may mean that the molten core collapses or splashes rapidly in the liquid state, while an excessively slow descent rate may mean that there is an unmelted solid core inside the molten core. To extract the integral area value, the system multiplies the angular interval between all adjacent data points within the decreasing resistance segment of the molten core by the average resistance value of those two data points, obtaining the trapezoidal area within each minute angular interval. Then, all these trapezoidal areas are summed item by item, and the sum is the integral area value of the decreasing segment relative to the angular axis. This integral area value characterizes the overall heat energy dissipated by the molten core during the contraction phase. It integrates information from both the width of the decreasing angle and the resistance level, indirectly reflecting the width of the plastic ring and the overall heat capacity of the molten pool during the contraction phase.

[0075] Finally, the four features mentioned above are assembled into a multidimensional original feature vector. The system extracts the average slope value of the rising segment of the weld nugget resistance, the peak resistance value of the rising segment of the weld nugget resistance, the falling rate value of the falling segment of the weld nugget resistance, and the integral area value of the falling segment of the weld nugget resistance, and arranges them in a fixed order to form a multidimensional original feature vector containing four dimensions. Each dimension in this vector reflects key information about the quality of weld nugget formation from different physical perspectives: the average slope value and the peak resistance value mainly characterize the energy accumulation characteristics during the welding heat input stage, while the falling rate value and the integral area value mainly characterize the morphological evolution characteristics during the solidification and shrinkage stage of the weld nugget. The four dimensions complement and corroborate each other, together forming a numerical feature set that can comprehensively describe the metallurgical quality of the current weld nugget, providing rich and physically meaningful original data support for the weighted fusion judgment of the quality assessment model in step seven.

[0076] S7. Call the preset welding quality quantification model, perform weighted fusion and normalization operations on the multidimensional original feature vector, and generate a real-time welding quality index.

[0077] Specifically, firstly, the internal structure and offline construction process of the welding quality quantification model are described. This model is essentially a multi-dimensional weight matrix and a set of normalized mapping parameters stored in the controller's memory. Its construction adopts a complete process of "standard sample calibration—destructive verification—weight regression." In the first stage of construction, the manufacturer selects standard specimens with the same material, surface condition, and thickness as the workpiece to be welded. Under the condition that the generator welding machine is in optimal working condition (i.e., the output voltage, current, and speed are all within the rated stable range), welding is performed according to the standard welding parameters verified by the process. For each standard specimen, the system completely executes the entire process from the first to the sixth step, extracting its corresponding multi-dimensional original feature vector, namely the average slope value of the rising segment of the weld nugget resistance, the peak resistance value, the falling rate value of the falling segment of the weld nugget resistance, and the integral area value, and storing these data in the calibration database. In the second stage of construction, destructive testing is performed on all the above standard specimens, specifically including tensile shear tests to determine the weld nugget diameter and breaking force, and metallographic microstructure analysis to observe the grain morphology of the fusion zone and the presence of porosity or cracks. Only the multidimensional original feature vectors corresponding to specimens with tensile strength meeting design specifications and dense, defect-free metallographic structure are included in the qualified positive sample set. In the third stage of construction, based on a large amount of data in the positive sample set, a multiple regression analysis method is used to determine the quantitative weights of the contribution of each of the four features to the final weld quality. The specific logic is as follows: statistical analysis is performed on the four feature values ​​of all qualified weld points in the positive sample set, and the correlation coefficient between each feature and the final destructive test result (such as tensile force) within its own physical fluctuation range is calculated. The feature with the larger the correlation coefficient is assigned a higher contribution weight in the model. This results in a multidimensional feature weight matrix containing four weight coefficients, which correspond to the average slope value, peak resistance value, descent rate value, and integral area value, respectively. Simultaneously, in the fourth stage of construction, the system calculates the sum of the above four feature values ​​of each qualified weld point in the positive sample set multiplied by its corresponding weight coefficient, and obtains a series of qualified benchmark fusion values. Then, the arithmetic mean and standard deviation of this series of qualified benchmark fusion values ​​are calculated. The mean is used as the central benchmark value for subsequent normalization, and the standard deviation is used as the benchmark dispersion to measure the dispersion of qualified samples. The above three data points—weight matrix, central benchmark value, and benchmark dispersion—together constitute a complete welding quality quantification model, which is burned into the controller of each generator-weld machine of the same model as fixed parameters.

[0078] Secondly, a weighted fusion operation is performed on the multidimensional original feature vector of the current solder joint. During the real-time detection phase, the system multiplies the four feature values ​​of the current solder joint extracted in step six—the average slope, peak resistance, descent rate, and integral area—one by one with the four corresponding weight coefficients stored in the model. After multiplication, the system sums these four products to obtain an unnormalized original fusion total value. The physical dimension of this original fusion total value is a weighted mixture of four different physical dimensions (resistance, resistance / angle, resistance*angle, etc.), and its magnitude is positively correlated with the combined effect of the overall heat input level and contraction morphology of the solder joint melt nugget. However, since the absolute value of this original fusion total value will exhibit systematic shifts across different welding batches and under different ambient temperatures, directly using this absolute value for judgment would result in a lack of universality and stability in the evaluation criteria.

[0079] Finally, a normalization operation is performed on the original total fusion value to generate a dimensionless real-time welding quality index. The system subtracts the center reference value stored in the model from the original total fusion value obtained above to obtain the deviation of the current weld point from the center position of the qualified sample. Subsequently, the system divides this deviation by the reference dispersion stored in the model, and the resulting quotient represents how many reference dispersions are used to measure the degree to which the current weld point's total fusion value deviates from the center of the qualified sample. To further limit the judgment range to a numerical interval that is easy to observe intuitively, the system performs a nonlinear amplitude limiting mapping on the quotient: when the quotient is within the range of negative two to positive three times the reference dispersion, the system linearly maps it to the integer interval of zero to one hundred, where the value of fifty corresponds to the center reference value position; when the quotient is lower than negative two times the reference dispersion or higher than positive three times the reference dispersion, the system clamps it to the boundary values ​​of zero and one hundred, respectively. After the above normalization and amplitude limiting mapping processing, the system finally outputs a dimensionless integer between zero and one hundred, which is the real-time welding quality index. The higher the index value, the further the overall characteristics of the current solder joint deviate from the high end of the qualified center, i.e., the heat input is too large; the lower the value, the further it deviates from the low end of the qualified center, i.e., the heat input is too small; and the closer the index is to fifty, the better the solder joint characteristics match the qualified standard sample that has undergone destructive testing.

[0080] S8. Compare the real-time welding quality index with the standard quality index range, determine whether the current weld quality is qualified, under-welded, or over-burned, and output the corresponding determination signal.

[0081] Specifically, firstly, the criteria and specific numerical boundaries for defining the standard quality index range are clearly defined. The standard quality index range is not arbitrarily set, but rather determined based on the normalized exponential distribution characteristics of the qualified positive sample set in step seven, combined with the actual allowable deviations in engineering practice. Specifically, the criteria are as follows: during the offline model construction phase, the real-time welding quality index corresponding to all weld points confirmed as qualified through destructive testing, after completing the entire process in step seven, approximates a normal distribution after statistical analysis. The system defines the coverage range between the 2.5 percentile and the 97.5 percentile of this normal distribution as the theoretical core range. Based on this, considering the slight fluctuations in the operating conditions of the on-site power generation welding machine and the differences in the habits of different operators, the system additionally extends the theoretical core range at both ends by an exponential step size corresponding to a benchmark dispersion as an engineering redundancy margin. After the above theoretical statistics and engineering margin extension, the finally determined standard quality index range has clear lower and upper boundary values. In the preferred setting of this embodiment, the lower boundary value corresponds to an exponential value of 45, and the upper boundary value corresponds to an exponential value of 75. Any weld joint whose real-time weld quality index falls within the range of 45 to 75 (inclusive) is considered to have statistical consistency with the qualified standard sample after destructive testing.

[0082] Secondly, a three-stage comparison is performed between the real-time welding quality index and the standard range. The system compares the real-time welding quality index generated in step seven with both the lower and upper boundary values ​​of the standard quality index range. This comparison process is executed at high speed by the numerical comparator hardware inside the controller, and the entire judgment process takes no more than one angle sampling cycle. The comparison logic is divided into three mutually exclusive cases: First case, if the real-time welding quality index is greater than or equal to the lower limit boundary value and less than or equal to the upper limit boundary value, that is, the index is within the closed interval of 45 to 75, then the system determines the current weld point to be a qualified product; Second case, if the real-time welding quality index is strictly less than the lower limit boundary value, that is, the index is lower than 45, then the system determines the current weld point to be a defective product, and specifically diagnoses the defect type as an under-welding defect caused by insufficient welding heat input or inadequate bonding of the fusion interface; Third case, if the real-time welding quality index is strictly greater than the upper limit boundary value, that is, the index is higher than 75, then the system determines the current weld point to be a defective product, and specifically diagnoses the defect type as an overheating defect caused by excessive welding heat input leading to overheating of the molten pool, splashing of molten iron, or a tendency to burn through.

[0083] Finally, based on the three-stage judgment results, the system synchronously executes graded output responses and protection linkage control. The system generates three different control signals from the judgment results, each responding through an independent output channel. For the first scenario (qualified), the system generates a qualification confirmation signal, which is sent to the green LED indicator light on the generator welding machine's control panel, keeping it constantly lit. Simultaneously, it drives a buzzer to emit a short beep, informing the operator that the current weld point is of acceptable quality and the next weld point can be welded. For the second scenario (under-welded) and the third scenario (over-burned), the system generates alarm signals carrying different defect type codes. These alarm signals are sent to the red LED indicator light and buzzer on the control panel, triggering continuous flashing of the red indicator light accompanied by a long beep, alerting the operator to a quality defect in the current weld point. More importantly, this alarm signal is simultaneously output via hardwired to the input terminal of the generator welding machine's protection control circuit. Upon receiving an alarm signal, the protection control circuit immediately executes two linked actions: First, it instantly cuts off the output contactor of the main welding circuit, ensuring timely termination of the current weld even at the end of the welding energizing process, preventing further deterioration of defects. Second, it triggers the preset compensation parameters stored internally in the controller. At the start of the next welding energizing cycle, it automatically adjusts the given value of the welding current or the set value of the welding energizing time based on the received defect type code. For example, if the previous weld was determined to be under-welded, the next weld will automatically increase the welding current by a preset compensation step; if the previous weld was determined to be over-burned, the next weld will automatically decrease the welding current by a preset compensation step, thus achieving adaptive point-by-point correction of weld quality. All the above judgment, display, alarm, and protection linkage actions are completed within ten milliseconds after the end of the current welding energizing cycle, truly achieving a closed-loop detection effect of "judgment upon welding, and control while judging."

[0084] The second embodiment of this application is as follows:

[0085] Please see Figure 3 This invention provides a welding quality inspection system for a generator welding machine based on a dynamic resistance curve, applicable to a welding quality inspection method for a generator welding machine based on a dynamic resistance curve as provided in the first embodiment, comprising:

[0086] The voltage sampling module and the current sampling module are connected in parallel and in series to the welding main circuit of the generator-weld machine, respectively, to synchronously collect the instantaneous terminal voltage signal at both ends of the workpiece and the instantaneous circuit current signal flowing through the welding circuit during welding power-on.

[0087] The rotor phase encoding unit is coaxially connected to the generator rotor of the generator welding machine and is used to output a phase pulse sequence corresponding to the rotor rotation angle;

[0088] The signal preprocessing unit, connected to the voltage sampling module, the current sampling module and the rotor phase encoding unit, is used to perform hardware filtering, analog-to-digital conversion and timestamp calibration on the acquired instantaneous terminal voltage signal and instantaneous loop current signal;

[0089] An equal-angle resampling unit, connected to the signal preprocessing unit, is used to convert a sampling sequence with a fixed time interval into a resampling terminal voltage sequence and a resampling loop current sequence with a fixed angle interval based on the phase pulse sequence.

[0090] The dynamic resistance curve reconstruction unit, connected to the equal-angle resampling unit, is used to perform point-by-point division and composite filtering on the resampling terminal voltage sequence and the resampling loop current sequence to generate a reference dynamic resistance curve.

[0091] The feature recognition and segmentation unit, connected to the dynamic resistance curve reconstruction unit, is used to automatically locate three key feature inflection points and segment the reference dynamic resistance curve into a contact resistance attenuation segment, a molten core resistance rising segment, and a molten core resistance falling segment.

[0092] The feature extraction unit, connected to the feature recognition and segmentation unit, is used to extract the average slope value and peak resistance value from the rising segment of the molten core resistor, and the falling rate value and integral area value from the falling segment of the molten core resistor, to generate a multi-dimensional original feature vector;

[0093] The quality assessment unit, built into the generator welding machine controller and connected to the feature extraction unit, is used to call the welding quality quantification model to perform weighted fusion and normalization on the multi-dimensional original feature vector to generate a real-time welding quality index, and compare the real-time welding quality index with the standard quality index range to determine the quality, and output the corresponding quality judgment signal to the display terminal and protection control loop.

[0094] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0095] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0096] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0097] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for inspecting the welding quality of a generator-welding machine based on dynamic resistance curves, characterized in that, Includes the following steps: During welding, the instantaneous terminal voltage signals at both ends of the workpiece, the instantaneous circuit current signal flowing through the welding circuit, and the phase pulse sequence of the generator rotor of the generator welding machine are synchronously collected at a fixed time sampling frequency. Hardware filtering and analog-to-digital conversion are performed on the acquired instantaneous terminal voltage signal and instantaneous loop current signal. Angle labels are added to the converted voltage and current data based on the phase pulse sequence to establish the correspondence between each sampling data point and the spatial angle of the generator rotor. Based on the angle label, the sampling sequence with a fixed time interval is converted into a resampled terminal voltage sequence and a resampled loop current sequence with a fixed angle interval. A point-by-point division operation is performed on the resampled terminal voltage sequence and the resampled loop current sequence to obtain the original dynamic resistance curve that varies with the angle. The original dynamic resistance curve is then subjected to sliding median filtering and zero phase shift low-pass filtering to generate a reference dynamic resistance curve. Based on the differential characteristics of the reference dynamic resistance curve, three key feature inflection points are automatically located, and the reference dynamic resistance curve is divided into a contact resistance attenuation segment, a molten core resistance rising segment, and a molten core resistance falling segment using the three key feature inflection points as dividing boundaries. The average slope value and peak resistance value are extracted from the rising segment of the molten core resistance, and the falling rate value and integral area value are extracted from the falling segment of the molten core resistance. The four feature values ​​together constitute a multidimensional original feature vector. A preset welding quality quantification model is invoked to perform weighted fusion and normalization operations on the multidimensional original feature vectors to generate a real-time welding quality index. The real-time welding quality index is compared with the standard quality index range to determine whether the current weld quality is qualified, under-welded, or over-burned, and the corresponding determination signal is output.

2. The welding quality inspection method for generator welding machines based on dynamic resistance curves as described in claim 1, characterized in that, During welding energization, instantaneous terminal voltage signals at both ends of the workpiece, instantaneous circuit current signals flowing through the welding circuit, and phase pulse sequences of the generator rotor of the welding machine are synchronously acquired at a fixed sampling frequency, including: The voltage sampling channel, current sampling channel, and rotor phase encoding acquisition channel are driven to operate synchronously by the master clock signal generated by the same high-stability clock source, ensuring that the sampling time deviation between channels is controlled within 1 microsecond; the phase pulse sequence is generated by an incremental photoelectric encoder coaxially connected to the generator rotor, and each phase pulse is triggered by a rising edge and the corresponding global timer count value is latched as a timestamp.

3. The welding quality inspection method for generator welding machines based on dynamic resistance curves as described in claim 1, characterized in that, The hardware filtering uses a second-order Butterworth active low-pass filter with a cutoff frequency of 3 kHz to filter out high-frequency switching noise; the analog-to-digital conversion uses a 16-bit successive approximation analog-to-digital converter to perform synchronous digitization at a sampling rate of 200 kilos per channel per second. The process of adding angle labels is as follows: taking the fixed angle interval between two adjacent rising edges of phase pulses as an angle interval unit, the timestamps of each voltage data and current data in the angle interval unit are converted into the corresponding angle positions in the angle interval unit according to the linear proportional distribution principle, thereby establishing a one-to-one correspondence between each sampled data point and the rotor spatial angle.

4. The welding quality inspection method for generator welding machines based on dynamic resistance curves as described in claim 1, characterized in that, Based on the angle label, the sampling sequence with a fixed time interval is converted into a resampled terminal voltage sequence and a resampled loop current sequence with a fixed angle interval, including: Using the base interval determined by the number of pulses per revolution of the encoder as the original angle interval unit, several equally spaced target angle grid points are generated within each original angle interval unit. The original sampled data falling within the original angle interval unit is interpolated and reconstructed using a piecewise cubic conformal interpolation algorithm. The reconstructed terminal voltage value and loop current value are obtained at each target angle grid point. Finally, the end-to-end are spliced ​​together to form a resampled terminal voltage sequence and a resampled loop current sequence with a fixed angle interval.

5. The welding quality inspection method for generator welding machines based on dynamic resistance curves as described in claim 1, characterized in that, The point-by-point division operation is performed one by one according to the same angle index position, and the set current points are invalidated and replaced by the current lower limit threshold; the sliding median filtering adopts a filter window with a preset angle width, sorts the resistance values ​​in the window and takes the median as the output value of the window center point, which is used to filter out the spikes caused by droplet short-circuit transition; the zero phase shift low-pass filtering adopts the method of cascading forward filtering and reverse filtering, which ensures that the position of the key feature inflection point on the angle axis does not shift while smoothing the residual ripple.

6. The welding quality inspection method for generator welding machines based on dynamic resistance curves as described in claim 1, characterized in that, The specific methods for locating the three key feature inflection points are as follows: The first feature inflection point is located at the peak point of solid-state resistance when the electrode contacts the workpiece at the beginning of the welding process, determined by finding the zero-crossing point where the first derivative changes from positive to negative within a preset first search window; the second feature inflection point is located at the turning point of the resistance rise rate after the rapid expansion of the weld nugget ends, determined by monitoring the position where the first derivative falls back from its maximum value to a preset proportional threshold within a second search window after the first feature inflection point; the third feature inflection point is located at the starting point of resistance decay as the weld nugget shrinks towards the liquid bridge after its growth is complete, determined by finding the zero-crossing point where the second derivative changes from negative to positive within a third search window after the second feature inflection point.

7. The welding quality inspection method for generator welding machines based on dynamic resistance curves as described in claim 1, characterized in that, The average slope value of the rising section of the molten core resistance is obtained by performing univariate linear regression trend analysis on all discrete data points in the section, and is used to characterize the rate of thermal accumulation of the molten core resistance. The peak resistance value is obtained by taking the maximum value of all resistance values ​​in the section, and is used to characterize the maximum resistance extreme value when the volume of the molten core expands to the limit. The falling rate value of the falling section of the molten core resistance is obtained by calculating the resistance difference between the starting point and the ending point of the section and dividing it by the angular width it crosses, and is used to characterize the speed at which the molten core contracts towards the liquid bridge. The integral area value is obtained by summing the areas of the trapezoids formed by adjacent data points in the section, and is used to characterize the overall heat dissipation characteristics of the molten core contraction stage.

8. The welding quality inspection method for generator welding machines based on dynamic resistance curves as described in claim 1, characterized in that, The construction process of the welding quality quantification model includes: selecting standard specimens for welding under standard process parameters, extracting multidimensional original feature vectors of qualified weld points as positive sample sets, determining the weight coefficients corresponding to the four features through multiple regression analysis to form a multidimensional feature weight matrix, and calculating the arithmetic mean of the fusion values ​​of the positive sample set as the central reference value and the standard deviation as the reference dispersion; the weighted fusion is to multiply the four feature values ​​of the current weld point by the corresponding weight coefficients and then sum them to obtain the original fusion total value; the normalization operation is to subtract the central reference value from the original fusion total value, divide it by the reference dispersion, and then perform nonlinear amplitude limiting mapping to obtain a dimensionless integer between zero and one hundred as the real-time welding quality index.

9. The welding quality inspection method for generator welding machines based on dynamic resistance curves as described in claim 1, characterized in that, The standard quality index range is determined based on the index distribution statistics of the qualified positive sample set combined with the engineering redundancy margin, and has a clear lower limit boundary value and an upper limit boundary value. When the real-time welding quality index falls within the standard quality index range, the current weld point is determined to be qualified and a qualified confirmation signal is output. When the real-time welding quality index is lower than the lower limit boundary value, the current weld point is determined to be under-welded and an under-welded alarm signal is output. When the real-time welding quality index is higher than the upper limit boundary value, the current weld point is determined to be over-burned and an over-burned alarm signal is output. The under-welded alarm signal and the over-burned alarm signal are synchronously output to the protection control circuit of the generator welding machine to trigger the welding main circuit to cut off and automatically correct the welding parameters of the next weld point.

10. A welding quality inspection system for a generator welding machine based on a dynamic resistance curve, used to implement the welding quality inspection method for a generator welding machine based on a dynamic resistance curve as described in claim 1, characterized in that, include: The voltage sampling module and the current sampling module are connected in parallel and in series to the welding main circuit of the generator-weld machine, respectively, to synchronously collect the instantaneous terminal voltage signal at both ends of the workpiece and the instantaneous circuit current signal flowing through the welding circuit during welding power-on. The rotor phase encoding unit is coaxially connected to the generator rotor of the generator welding machine and is used to output a phase pulse sequence corresponding to the rotor rotation angle; The signal preprocessing unit, connected to the voltage sampling module, the current sampling module and the rotor phase encoding unit, is used to perform hardware filtering, analog-to-digital conversion and timestamp calibration on the acquired instantaneous terminal voltage signal and instantaneous loop current signal; An equal-angle resampling unit, connected to the signal preprocessing unit, is used to convert a sampling sequence with a fixed time interval into a resampling terminal voltage sequence and a resampling loop current sequence with a fixed angle interval based on the phase pulse sequence. The dynamic resistance curve reconstruction unit, connected to the equal-angle resampling unit, is used to perform point-by-point division and composite filtering on the resampling terminal voltage sequence and the resampling loop current sequence to generate a reference dynamic resistance curve. The feature recognition and segmentation unit, connected to the dynamic resistance curve reconstruction unit, is used to automatically locate three key feature inflection points and segment the reference dynamic resistance curve into a contact resistance attenuation segment, a molten core resistance rising segment, and a molten core resistance falling segment. The feature extraction unit, connected to the feature recognition and segmentation unit, is used to extract the average slope value and peak resistance value from the rising segment of the molten core resistor, and the falling rate value and integral area value from the falling segment of the molten core resistor, to generate a multi-dimensional original feature vector; The quality assessment unit, built into the generator welding machine controller and connected to the feature extraction unit, is used to call the welding quality quantification model to perform weighted fusion and normalization on the multi-dimensional original feature vector to generate a real-time welding quality index, and compare the real-time welding quality index with the standard quality index range to determine the quality, and output the corresponding quality judgment signal to the display terminal and protection control loop.

Images