High-frequency welded pipe production quality on-line monitoring intelligent diagnosis system

Abstract

The invention relates to the technical field of high-frequency welding equipment, and discloses a high-frequency welded pipe production quality on-line monitoring intelligent diagnosis system, which comprises a data acquisition unit, which is configured to synchronously acquire a high-frequency voltage waveform and a high-frequency current waveform generated by a high-frequency welding loop under the common excitation of a main welding frequency signal and an injected probe signal; the feature extraction unit is connected with the data acquisition unit, and is configured to extract at least one passive electromagnetic fingerprint and one active response fingerprint based on the waveform, and fuse the at least one passive electromagnetic fingerprint and the active response fingerprint to generate a welding state vector; and the intelligent diagnosis unit is connected with the feature extraction unit and is configured to compare the welding state vector with a preset health model so as to judge the current welding quality of the high-frequency welded pipe. According to the method, through an active and passive information fusion strategy, the welding state can be represented more comprehensively, and the detection sensitivity and diagnosis accuracy of early weak defects are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-frequency welding equipment, in particular to an intelligent diagnosis system for online monitoring of high-frequency welded pipe production quality. BACKGROUND

[0002] High-frequency induction welding is one of the mainstream processes for producing high-quality straight seam welded pipes in the current industrial field. This process is widely used in many industries such as petroleum, natural gas, construction, and mechanical manufacturing due to its fast welding speed, narrow heat-affected zone, and high production efficiency. In the high-frequency welding process, the edges of the steel strip are rapidly heated to a molten or plastic state by high-frequency induction current when passing through the V-shaped guide zone, and then solid-phase welding is achieved under the strong pressure of the extrusion roller, forming a continuous and dense weld.

[0003] The quality of the weld directly determines the overall strength, sealing performance, and service life of the welded pipe, and is the core control link in the entire production process. However, high-frequency welding is a complex dynamic process involving electromagnetic, thermal, and force multi-physical field strong coupling, and is easily affected by various uncertain factors such as steel strip material fluctuations, equipment state changes, and environmental disturbances, thereby inducing various welding defects such as cold welding, overburning, inclusions, and cracks.

[0004] Traditional welded pipe quality control methods rely mainly on offline detection, such as destructive testing such as flattening and flaring of finished pipes, or non-destructive testing such as ultrasonic and eddy current testing. Although these methods can assess whether the final product is acceptable, they are essentially "post-inspection" and have significant lag. When defects are detected, a large number of substandard pipes have already been produced, resulting in significant material waste and economic losses, and providing no effective information for real-time adjustment of process parameters and prevention of defect occurrence.

[0005] In order to overcome the disadvantages of offline detection, the industry has developed various online monitoring technologies. For example, the temperature distribution of the weld area is monitored by an infrared thermal imager, or the output voltage, current, power and other macro electrical parameters of the welding power supply are directly monitored. However, these existing online monitoring methods still have obvious limitations. The method based on infrared thermal imaging is extremely susceptible to environmental factors such as splashing, water vapor and smoke in the welding site, and the change of the emissivity of the metal surface will also affect the accuracy of temperature measurement. It can only capture the thermal appearance of the process. While monitoring the macro electrical parameters of the welding power supply, although it can reflect the stability of the overall energy supply, the sensitivity is often insufficient for the local and transient physical process changes that occur in the V-shaped convergence zone, which is the core area of welding, which is crucial for defect formation. The complexity of the welding process determines that single-dimensional information monitoring cannot fully and accurately depict its true state. A small abnormality in one parameter may be masked by changes in other parameters, leading to missed or misdiagnosed defects.

[0006] Therefore, there is an urgent need in the art for an online monitoring and diagnosis technology that can deeply understand the physical nature of the welding process, provide real-time, sensitive and multi-dimensional information, in order to achieve early warning and accurate diagnosis of welding quality abnormalities, and thus provide technical support for intelligent quality control of high-frequency welded pipe production. SUMMARY

[0007] The technical problem to be solved by the present application is that the existing high-frequency welded pipe quality monitoring technology has the limitations of insufficient real-time performance, inability to effectively detect internal defects in the weld, and difficulty in accurately diagnosing the root cause of the defects.

[0008] To solve the above technical problems, the present application provides the following technical solutions:

[0009] The first aspect of the present application provides an online monitoring and intelligent diagnosis system for high-frequency welded pipe production quality, which comprises:

[0010] a data acquisition unit configured to synchronously acquire a high-frequency voltage waveform and a high-frequency current waveform in a high-frequency welding circuit, the high-frequency voltage waveform and the high-frequency current waveform being generated by the high-frequency welding circuit under the joint excitation of a main welding frequency signal and an injected probe signal;

[0011] a feature extraction unit connected to the data acquisition unit and configured to extract at least one passive electromagnetic fingerprint and one active response fingerprint based on the high-frequency voltage waveform and the high-frequency current waveform, and to generate a welding state vector by fusion; the passive electromagnetic fingerprint includes at least one of instantaneous complex impedance and / or harmonic energy ratio, and the active response fingerprint is a response feature of the high-frequency welding circuit to the probe signal;

[0012] An intelligent diagnosis unit, connected with the feature extraction unit, is configured to compare the welding state vector with a preset health model to determine the current high-frequency welded pipe welding quality.

[0013] Preferably, the system further comprises:

[0014] A perturbation signal injection module, coupled with the high-frequency welding loop, is configured to inject the probe signal into the high-frequency welding loop; the frequency of the probe signal is different from the fundamental frequency of the main welding frequency signal and its integer harmonic frequency.

[0015] In a specific embodiment, when the feature extraction unit extracts the passive electromagnetic fingerprint, it is configured to:

[0016] Calculate the ratio between the high-frequency voltage waveform v(t) and the high-frequency current waveform i(t) to obtain the instantaneous complex impedance Z eq (t), whose mathematical expression is:

[0017] Z eq (t) = R eq (t) + jX eq (t);

[0018] Wherein, R eq (t) is an equivalent resistance representing energy dissipation, and X eq (t) is an equivalent reactance representing electromagnetic energy storage;

[0019] And / or

[0020] Perform spectral analysis on the high-frequency current waveform to obtain the time-frequency spectrum I(ω, t) to obtain the harmonic energy ratio HER n (t) representing the degree of nonlinearity of the welding process, whose mathematical expression is:

[0021]

[0022] Wherein, f w is the fundamental frequency of the main welding frequency signal, and n is the harmonic order.

[0023] In a specific embodiment, when the feature extraction unit extracts the active response fingerprint, it is configured to:

[0024] Using a digital phase-locked amplification algorithm, the response component caused by the probe signal is locked and extracted from the high-frequency current waveform and / or high-frequency voltage waveform to obtain the active response fingerprint.

[0025] The active response fingerprint includes the amplitude attenuation A p (t) and / or phase shift φ p(t), whose mathematical expression is:

[0026]

[0027] φ p (t) = atan2(C Q (t), C I (t));

[0028] where C I (t) and C Q (t) are the DC components obtained by mixing the input signal with in-phase and quadrature reference signals and low-pass filtering, respectively.

[0029] Preferably, the intelligent diagnostic unit is configured to compare the welding state vector with a pre-set health model when comparing the welding state vector with the pre-set health model.

[0030] The probability distribution model established based on a plurality of historical welding state vectors collected when producing qualified welded pipes is taken as the health model.

[0031] The Mahalanobis distance D M (t) between the current welding state vector S(t) and the health model is calculated to quantify the degree of deviation, and the welding quality is determined based on the degree of deviation. The mathematical expression of the Mahalanobis distance is:

[0032]

[0033] where μ H and∑ H are the mean vector and covariance matrix of the health model, respectively.

[0034] Further, the intelligent diagnostic unit further comprises a defect evolution path tracking module configured to store a time sequence of the welding state vectors to constitute an evolution path, and identify the mode of defects based on the dynamic properties of the evolution path.

[0035] In a specific embodiment, the defect evolution path tracking module is configured to:

[0036] analyze the geometric and dynamic properties of the evolution path in the multi-dimensional feature space, including the speed, curvature and direction change of the trajectory, to classify the mode of defects into one of a point defect, a linear defect or a periodic defect.

[0037] Preferably, the system is further configured to:

[0038] When the defect evolution path tracking module identifies a preset defect mode, the perturbation signal injection module is triggered to adjust or inject a specific probe signal sequence to target the defect mode for targeted exploration and confirmation.

[0039] In one specific embodiment, the data acquisition unit is further configured to synchronously acquire at least one external physical signal, the external physical signal including at least one of thermal field information of the weld area or geometric information of the weld reinforcement;

[0040] The feature extraction unit is further configured to fuse the features extracted from the external physical signal into the welding state vector.

[0041] The second aspect of the present application provides an intelligent diagnosis method for online monitoring of high-frequency pipe welding production quality, which comprises the following steps:

[0042] The acquisition step comprises synchronously acquiring high-frequency voltage and current waveforms generated by the high-frequency welding circuit under the joint excitation of the main welding frequency signal and the injected probe signal;

[0043] The feature extraction step comprises extracting at least one passive electromagnetic fingerprint and one active response fingerprint based on the high-frequency voltage and current waveforms, and fusing to generate a welding state vector; the passive electromagnetic fingerprint includes at least one of instantaneous complex impedance and / or harmonic energy ratio, and the active response fingerprint is a response feature of the high-frequency welding circuit to the probe signal;

[0044] The diagnosis step comprises comparing the welding state vector with a preset health model to determine the current high-frequency pipe welding quality.

[0045] The present application provides an intelligent diagnosis system for online monitoring of high-frequency pipe welding production quality, which has the following beneficial effects:

[0046] 1. The present application injects a probe signal of a specific frequency into the high-frequency welding circuit through the perturbation signal injection module, and extracts passive electromagnetic fingerprints and active response fingerprints by the feature extraction unit. Since the active response fingerprint is extracted from the response to the known probe signal through digital lock-in amplification algorithm, this process can effectively suppress the broadband noise from the main welding frequency signal and environmental electromagnetic interference, thereby significantly improving the detection sensitivity of the system to the subtle changes in the material electrical properties of the weld area, enhancing the ability to identify early or small defects, and improving the signal-to-noise ratio and robustness of the diagnosis.

[0047] 2、The application can distinguish the defect mode into different types such as point, linearity or periodicity by analyzing the speed, curvature and other dynamic properties of the path in the feature space, which can provide more technical information for tracing the root cause of the defect (for example, is it a random material problem or a systematic equipment problem), thereby improving the depth and accuracy of diagnosis.

[0048] 3、The application can realize collaborative diagnosis refinement by establishing a feedback mechanism between the defect evolution path tracking module and the perturbation signal injection module. When the system identifies an uncertain defect mode, it can actively adjust or inject specific probe signal sequences for targeted exploration, and confirm or exclude the preliminary diagnosis according to the response results. This closed-loop workflow of discovery-exploration-confirmation avoids the misjudgment caused by relying on single passive monitoring, thereby effectively reducing the false positive rate of the diagnosis results and improving the overall confidence and reliability of the system output conclusion. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The structural block diagram of the high-frequency welded pipe production quality online monitoring intelligent diagnosis system of an embodiment of the application;

[0050] Figure 2 The typical frequency spectrum diagram of the mixed signal collected in the embodiment of the application;

[0051] Figure 3 The flowchart of the high-frequency welded pipe production quality online monitoring intelligent diagnosis method of an embodiment of the application;

[0052] Figure 4 The Mahalanobis distance time series diagram displayed on the system monitoring interface in the embodiment of the application.

[0053] Among them, 10, perturbation signal injection module; 20, data acquisition unit; 30, feature extraction unit; 40, intelligent diagnosis unit. DETAILED DESCRIPTION

[0054] The technical solutions of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0055] Refer to the drawings Figure 1 , Figure 1is a structural block diagram of an intelligent diagnosis system for online monitoring of high-frequency welded pipe production quality according to an embodiment of the present application. The system is used for online monitoring and diagnosis of the welding production process of high-frequency welded pipes.

[0056] In one specific embodiment, the system comprises a perturbation signal injection module 10, a data acquisition unit 20, a feature extraction unit 30, and an intelligent diagnosis unit 40.

[0057] The perturbation signal injection module 10 is electrically coupled with the high-frequency welding circuit for applying a probe signal to the circuit. The sensing components of the data acquisition unit 20 are coupled with the high-frequency welding circuit for acquiring the electrical parameter signals of the circuit. The output of the data acquisition unit 20 is electrically connected with the input of the feature extraction unit 30. The output of the feature extraction unit 30 is electrically connected with the input of the intelligent diagnosis unit 40. In one embodiment, a control output of the intelligent diagnosis unit 40 is connected with a control input of the perturbation signal injection module 10 to form a feedback path.

[0058] Specifically, the function of the perturbation signal injection module 10 is to generate a probe signal of a specific frequency and inject it into the high-frequency welding circuit. The frequency of the probe signal is set to be different from the fundamental frequency of the main welding frequency signal and its integer harmonic frequencies of the high-frequency welding circuit, to ensure its distinguishability in the frequency spectrum.

[0059] The function of the data acquisition unit 20 is to synchronously acquire the high-frequency voltage waveform and the high-frequency current waveform generated by the high-frequency welding circuit under the joint excitation of the main welding frequency signal and the probe signal. The unit converts the acquired analog waveform signals into digital signal streams and outputs them to the subsequent units for processing.

[0060] The function of the feature extraction unit 30 is to receive the digital signal streams from the data acquisition unit 20. The unit performs algorithm processing to extract at least one passive electromagnetic fingerprint and one active response fingerprint. The passive electromagnetic fingerprint is based on the analysis result of the response to the main welding frequency signal, and the active response fingerprint is based on the analysis result of the response to the probe signal. The unit further combines and normalizes the extracted multiple fingerprint features to generate a multi-dimensional welding state vector.

[0061] The function of the intelligent diagnosis unit 40 is to receive the welding state vector from the feature extraction unit 30. The unit internally stores a preset health model, which is a state vector probability distribution model established based on historical qualified welded pipe production data. The unit quantifies the deviation degree of the current welding state by mathematically comparing the real-time received welding state vector with the health model, and determines the welding quality of the current welded pipe based on the deviation degree, and finally outputs the determination result.

[0062] Reference to the drawings Figure 1 The specific structure and working mode of the perturbation signal injection module 10 in the embodiment of the present application will be described in detail.

[0063] The perturbation signal injection module 10 is configured to generate a probe signal with stable frequency and amplitude, and apply it to the high-frequency welding circuit.

[0064] In one embodiment, the perturbation signal injection module 10 is inductively coupled with the inductor coil or output bus of the high-frequency welding circuit through a high-frequency isolation transformer. The secondary winding of the isolation transformer is connected in series or parallel in the high-frequency welding circuit, and the probe signal is superimposed on the main welding current. This coupling mode provides electrical isolation to prevent the high-power main welding circuit from impacting the low-voltage control circuit inside the perturbation signal injection module 10.

[0065] In another embodiment, the perturbation signal injection module 10 can be directly connected to the control system of the high-frequency welding power supply, and the characteristics of the probe signal are indirectly superimposed into the output voltage or current waveform of the power supply by modulating the control signal (e.g., pulse width modulation signal) that drives the high-frequency inverter power device.

[0066] The internal part of the perturbation signal injection module 10 includes a digital frequency synthesizer (DDS), a linear power amplifier, and a control processor.

[0067] The control processor sets the frequency and amplitude of the probe signal according to the preset parameters or instructions from the intelligent diagnosis unit 40. The digital frequency synthesizer generates a high-precision low-voltage sinusoidal signal accordingly. The signal is then power amplified by the linear power amplifier, and the output power is controlled within a predetermined range (e.g., 1 to 10 watts), which is much lower than the kilowatt or megawatt main output power of the high-frequency welding power supply, ensuring that the injection of the probe signal does not affect the macroscopic welding heat input and welding quality.

[0068] The frequency f p of the probe signal is selected according to the following principles to ensure the separability of its response signal:

[0069]

[0070] Where f w is the fundamental frequency of the main welding frequency signal, and n is a positive integer.

[0071] This frequency selection principle can be intuitively understood by referring to the following Figure 2 The following Figure 2 is a typical frequency spectrum diagram of the mixed signal collected in the embodiment of the present application. As can be seen from the diagram, the energy of the signal is mainly concentrated in the main welding frequency f w(e.g. 400 kHz as shown in the figure) and its integer harmonic frequencies (e.g. 2nd harmonic 2f w , 3rd harmonic 3f w , etc.) form several high-amplitude spectral peaks.

[0072] The key of the present invention is that the selected probe signal frequency f p (e.g. 50 kHz as shown in the figure) is intentionally set in the "spectral gap" between these main harmonic spectral peaks.

[0073] This frequency setting principle avoids the spectrum of the probe signal and its response signal from overlapping with the spectrum of the main welding frequency signal and its harmonics, thus ensuring that the active response fingerprint generated by the probe signal excitation only can be unambiguously locked and extracted in the feature extraction unit 30 later by digital lock-in amplification algorithms, etc.

[0074] In the regular monitoring mode, the perturbation signal injection module 10 outputs a continuous, single-frequency sinusoidal wave as the probe signal. In addition, the control processor of the perturbation signal injection module 10 is also configured to receive control instructions from the intelligent diagnostic unit 40. When receiving specific instructions, the perturbation signal injection module 10 can switch the working mode to output a specially modulated probe signal sequence, for example, to perform a small range of frequency sweep (linear sweep signal), or to output a signal with a specific amplitude envelope for targeted exploration of specific defect modes.

[0075] Referring to the attached Figure 1 , the sensor array included in the data acquisition unit 20 in the embodiment of the present invention is described in detail. The sensor array is used to acquire multi-physical field measurement data related to the high-frequency welded pipe welding process.

[0076] The data acquisition unit 20 includes a sensor group for measuring electromagnetic signals and a sensor group for measuring external physical signals.

[0077] The electromagnetic signal sensor group is configured to non-invasively measure high-frequency voltage and current in the high-frequency welding circuit.

[0078] In one embodiment, the current sensor is a high-bandwidth Rogowski coil that is wrapped around the conductor connecting the high-frequency welding power supply and the induction coil. The bandwidth of the Rogowski coil is set to cover the range from the main welding frequency f w to the probe signal frequency f p and its response harmonics, for example, its -3dB bandwidth is not less than 5MHz. The voltage sensor is a high-voltage differential probe whose probing terminals are connected in parallel to the two ends of the induction coil. The probe has a bandwidth matching that of the current sensor, and a high common-mode rejection ratio (CMRR) to suppress the influence of common-mode noise voltage on differential-mode voltage measurement.

[0079] The external physical signal sensor group is configured to collect thermal field and geometric information directly related to the weld formation. In one embodiment, the thermal field information is acquired by a multi-point infrared pyrometer array. The array's multiple probes are precisely aimed at the heating area of the apex of the V-shaped region of the weld and the edges of the two side strips. Each pyrometer is of the narrow-band spectral response type to reduce the influence of the fluctuation of the emissivity caused by the change of the material surface state on the temperature measurement results, and its response time is set to be millisecond level or less to capture the temperature transient change in the welding process.

[0080] The geometric information is acquired by a laser profilometer installed at a position after the welding extrusion roller and before the weld reinforcement cutting station. The instrument uses the principle of laser triangulation to project a laser line across the weld area and capture its profile by an image sensor. The instrument works continuously at a preset sampling frequency (for example, 2000 profile lines are collected per second) to measure the geometric characteristics of the weld reinforcement such as height, width and cross-sectional shape in real time. Its measurement resolution is set at the micron level to detect the slight deviation of the weld geometry.

[0081] Further, the synchronization and acquisition hardware inside the data acquisition unit 20 in the embodiment of the present application is described in detail. The hardware is used to ensure the strict alignment of the multiple heterogeneous signals from the sensor array on the time reference, and to convert them into digital data with high fidelity.

[0082] In one embodiment, a network synchronization scheme based on the Precision Time Protocol (PTP, IEEE 1588) is used inside the data acquisition unit 20. The scheme includes an industrial Ethernet switch supporting the PTP protocol and one or more distributed data acquisition modules. A dedicated device in the switch or network acts as the master clock, broadcasting high-precision time synchronization messages to all data acquisition modules in the network. Each data acquisition module acts as a slave clock, receiving the message and continuously calibrating its local internal clock, so that the deviation of the clock of all modules from the master clock is controlled within 1 microsecond. When the modules collect sensor signals, a timestamp based on the high-precision synchronization clock is attached to each sampling point or data packet.

[0083] In another embodiment, the synchronization is achieved through a central clock distribution system. The system includes a highly stable master oscillator, such as a temperature compensated crystal oscillator (TCXO), for generating a uniform sampling clock signal and a periodic synchronization trigger signal. These signals are distributed to the external clock input and external trigger input of each data acquisition device (e.g., multiple data acquisition cards installed in the same industrial computer or PXI chassis) through impedance-matched shielded coaxial cables. This approach forces all data acquisition channels to sample at the same clock edge, thereby achieving synchronization at the hardware level.

[0084] The acquisition hardware of the data acquisition unit 20 is a multi-channel, synchronized sampling data acquisition card or module. To ensure complete capture of all information contained in the high-frequency voltage and current waveforms, including the main welding frequency signal, the probe signal and its response, each channel of the acquisition hardware has a sampling rate of no less than 10 MS / s (100 million samples per second) and a vertical resolution of no less than 16 bits. This resolution ensures that small amplitude signal components can be accurately quantized against the background of large amplitude main signals. The acquisition card transmits the converted digital data with high-precision time stamps to the memory of the host computer through a high-speed bus (e.g., PCIe) for subsequent processing by the feature extraction unit 30.

[0085] Referring to the accompanying drawings Figure 1 The specific process of extracting the passive electromagnetic fingerprint by the feature extraction unit 30 in the embodiment of the present application is described in detail. The passive electromagnetic fingerprint is a feature extracted based on the response of the high-frequency welding circuit to the main welding frequency signal, and this process does not depend on the injected probe signal.

[0086] In one embodiment, the feature extraction unit 30 receives the synchronized high-frequency voltage waveform digital signal v(t) and high-frequency current waveform digital signal i(t) from the data acquisition unit 20.

[0087] In order to obtain the instantaneous complex impedance Z eq (t), the feature extraction unit 30 first performs Hilbert transform on the received v(t) and i(t) respectively to construct their corresponding analytic signals. The analytic signal v a (t) of the voltage and the analytic signal i a (t) of the current are expressed as:

[0088]

[0089]

[0090] wherein, denotes the Hilbert transform operator, and j is the imaginary unit. By calculating the ratio of the two analytic signals, the instantaneous complex impedance Z eq(t):

[0091]

[0092] The real part R eq (t) of the calculation result eq (t) is the equivalent resistance, whose value change is associated with the joule heat energy dissipation of the welding area. The imaginary part X eq (t) is the equivalent reactance, whose value change is associated with the change of the molten pool geometry and the material permeability.

[0093] In order to obtain the harmonic energy ratio HER n (t), the feature extraction unit 30 performs a short-time Fourier transform (STFT) on the high-frequency current waveform digital signal i(t). The process includes:

[0094] Divide the continuous i(t) signal stream into short-time analysis windows with a certain overlap rate;

[0095] Apply a window function (for example, Hanning window) to the signal in each analysis window to suppress spectral leakage;

[0096] Perform a fast Fourier transform (FFT) on the windowed signal segment to obtain the discrete time-frequency spectrum I(ω, t) corresponding to the time period.

[0097] The feature extraction unit 30 extracts the fundamental welding frequency f w from the calculated time-frequency spectrum I(ω, t), the energy of the fundamental component, and the energy of each harmonic (for example, n = 2, 3, 5) component, and calculates the ratio thereof. The nth harmonic energy ratio HRE n (t) is calculated as follows:

[0098]

[0099] The value change of this feature reflects the change of the degree of system nonlinearity caused by physical phenomena such as arc, plasma discharge, etc. during the welding process. The feature extraction unit 30 repeats this calculation for each analysis window over time, thereby generating a series of passive electromagnetic fingerprint feature values that change over time.

[0100] Further, the specific process of the feature extraction unit 30 in the embodiment of the present application to extract the active response fingerprint is described in detail. The active response fingerprint is a feature extracted based on the response of the high-frequency welding circuit to the known probe signal. This process uses a digital lock-in amplification algorithm to separate the weak probe signal response component from the background containing strong main frequency signals and wideband noise.

[0101] The feature extraction unit 30 receives the synchronized high-frequency current waveform digital signal i(t) from the data acquisition unit 20, and learns the frequency f of the current probe signal from the perturbation signal injection module 10 or the system preset parameters p .

[0102] The first step of the algorithm is to generate two mutually orthogonal digital reference signals, i.e. the in-phase reference signal s ref，I (t) and the quadrature reference signal s ref，Q (t). The frequencies of the two signals are the same as the probe signal frequency f p , and the mathematical expressions are as follows:

[0103] s ref，I (t) = sin(2πf p t);

[0104] s ref，Q (t) = cos(2πf p t);

[0105] The second step of the algorithm is mixing. The feature extraction unit 30 point-by-point multiplies the input current signal i(t) with the above two reference signals to obtain two mixed signals. This operation down-converts the components with the frequency f p in the input signal to direct current (0 Hz) and twice frequency (2f p ).

[0106] The third step of the algorithm is low-pass filtering. The feature extraction unit 30 passes the two mixed signals through a digital low-pass filter to filter out the twice frequency components and other high-frequency noise components, and only retains the direct current components. This process is equivalent to integration operation in the time period T int , and finally obtains the in-phase direct current component C I (t) and the quadrature direct current component C Q (t). The mathematical expressions are as follows:

[0107]

[0108] wherein the selection of the integration period T int is related to the cutoff frequency of the low-pass filter, and the value of T int should be much larger than the period 1 / f p of the probe signal.

[0109] The fourth step of the algorithm is to calculate the amplitude and phase. The feature extraction unit 30 calculates the amplitude attenuation A p (t) and the phase shift φ p (t) of the probe signal response based on the obtained in-phase and quadrature direct current components. The two together constitute the active response fingerprint. The mathematical expressions are as follows:

[0110]

[0111] φ p (t) = atan2(C Q (t), C I (t));

[0112] where atan2 is the two-argument arctangent function.

[0113] The feature extraction unit 30 continuously performs the above steps to generate the time-varying A p (t) and φ p (t) feature value sequences. In another embodiment, only the high-frequency voltage waveform v(t) can be used as input, or both v(t) and i(t) can be used for computation.

[0114] After computing the passive electromagnetic fingerprint and the active response fingerprint, the feature extraction unit 30 performs a fusion step to construct a unified weld condition vector. This vector is the mathematical input for the subsequent intelligent diagnosis unit 40 to analyze.

[0115] First, the feature extraction unit 30 collects all the computed feature values at the same time point t.

[0116] In one embodiment, the collection includes the equivalent resistance R eq (t), the equivalent reactance X eq (t), and the second harmonic energy ratio HER2(t) in the passive electromagnetic fingerprint, and the amplitude attenuation A p (t) and the phase shift φ p (t) in the active response fingerprint. In embodiments that include external physical signal sensors, features extracted from thermal field information or geometric information (e.g., weld peak temperature, reinforcement width) are also included in this collection.

[0117] Since each feature in the collection has different physical units and numerical ranges (e.g., ohm, radian, dimensionless ratio), directly combining them would cause some features to have unbalanced weights in the subsequent mathematical model. To this end, the feature extraction unit 30 normalizes each feature sequence. In one specific embodiment, the Z-score standardization method is used. For any feature value x i , its normalized value x' i is computed as follows:

[0118]

[0119] where μ i is the mean of a large number of samples collected under historical normal production conditions, and σ iThis is its corresponding standard deviation. These two parameters (μ) i ,σ i The statistic is pre-calculated and stored in the system as a baseline statistic for that feature.

[0120] Finally, the feature extraction unit 30 collects all the standardized feature values ​​x′ at time point t. i Arranged in a predetermined order, they form a multi-dimensional column vector. This vector is defined as the welding state vector S(t) at that moment. Its structure is as follows:

[0121]

[0122] This welding state vector S(t) provides a dimensionless, multi-dimensional quantitative description of the current welding process and is transmitted to the intelligent diagnostic unit 40 as the final output of the feature extraction unit 30.

[0123] See attached document Figure 1 The specific functions of the intelligent diagnostic unit 40 in this embodiment of the invention will be described in detail. The core function of this unit is to perform abnormal detection based on a preset health model.

[0124] The intelligent diagnostic unit 40 receives a continuous welding state vector sequence S(t) from the feature extraction unit 30. This unit internally stores a health model, which is a mathematical description of the statistical distribution of welding state vectors during the production of qualified welded pipes. The model is built in an offline training phase. Specifically, firstly, within a production cycle confirming the production of defect-free welded pipes, a large number of welding state vector samples are collected, forming a health state sample set {S}. H}. Then, the mean vector μ of this sample set is calculated. H The sum of the covariance matrix ∑ H These two factors together constitute the health model in this embodiment. The calculation formula is as follows:

[0125]

[0126] Where N is the total number of samples in the healthy state sample set, and S H，k It is the k-th health state vector in the sample set. The mean vector μ H The covariance matrix ∑ represents the central location of the health status. H It describes the fluctuation range of each characteristic component under healthy conditions and their linear correlation with each other.

[0127] During the online diagnostic phase, the intelligent diagnostic unit 40 calculates the Mahalanobis distance D between each real-time received welding state vector S(t) and the established health model. M(t). Mahalanobis distance is an effective distance metric in multi-dimensional space, which takes into account the correlation between features and is not sensitive to the scale of features. Its formula is as follows:

[0128]

[0129] where S(t) is the current welding state vector, μ H and∑ H are the mean vector and covariance matrix of the healthy model, respectively, is the inverse of the covariance matrix. The calculated D M (t) is a scalar value, which quantifies the statistical distance of the current state from the center of the healthy state.

[0130] The intelligent diagnosis unit 40 compares the calculated Mahalanobis distance D M (t) with a preset decision threshold D threshold . The threshold is determined based on the Mahalanobis distance distribution of the healthy sample set (theoretically subject to chi-square distribution) and according to a preset confidence level (e.g. 99.7%). If D M (t) > D threshold , the current welding state is determined to be abnormal, and the system generates an abnormal state signal; if D M (t) ≤ D threshold , the current welding state is determined to be normal. The determination result is output to the subsequent defect evolution path tracking module for further analysis, or directly displayed on the human-machine interaction interface.

[0131] Further, the function of the defect evolution path tracking module inside the intelligent diagnosis unit 40 is described in detail. This module is used for in-depth dynamic analysis of the welding state sequence detected by the anomaly to identify the root mode of the defect.

[0132] The module receives and stores the continuous welding state vectors S(t) from the previous processing steps. In one embodiment, the module maintains a first-in-first-out (FIFO) ring buffer inside, which is used to store the last M welding state vectors. This time-ordered vector set {S(t-M+1),..., S(t)} forms a discrete trajectory in the multi-dimensional feature space, which is the defect evolution path.

[0133] The module quantitatively calculates the geometric and dynamic properties of the evolution path. In one embodiment, the module calculates the speed and curvature of the path. The instantaneous velocity vector v(t) of the path at time t is approximated by the first-order difference of the state vector:

[0134]

[0135] where Δt is the time interval between two consecutive state vectors. The magnitude of this velocity vector, i.e. the speed ||v(t)||, characterizes the intensity of the welding state change.

[0136] The curvature of the path characterizes the degree of its bending, reflecting the rate of change of the direction of the state change. In one embodiment, the module first computes the acceleration vector a(t) of the path, i.e. the first order difference of the velocity vector:

[0137]

[0138] Subsequently, the curvature of the path is computed based on the velocity vector and the acceleration vector. In another embodiment, the curvature is approximated by computing the rate of change of the angle between consecutive velocity vectors v(t) and v(t-1).

[0139] Based on the time series of the above dynamic attributes, the module classifies the defect pattern:

[0140] If the module detects an isolated, high-amplitude pulse of the speed ||v(t)|| and the curvature in a short time, followed by a rapid return to the baseline level, it classifies this event as a point defect. This corresponds to a short-lived, sharp excursion of the evolution path from the healthy region and a quick return.

[0141] If the module detects that the path migrates from the healthy region to another quasi-stable abnormal region at a relatively high speed, and continues to run in this region at a relatively low speed, it classifies this event as a linear defect. This corresponds to a persistent shift in the welding state.

[0142] If the module detects that the path presents a periodic trajectory in the feature space, such as a closed loop or oscillation, and its dynamic attributes such as speed and curvature also present corresponding periodic changes, it classifies this event as a periodic defect. The module can determine the characteristic frequency of this periodic change by performing a spectral analysis on the time series of one or more components of the state vector.

[0143] Further, the collaborative diagnosis refinement mechanism inside the intelligent diagnosis unit 40 in the embodiment of the present application is described in detail. This mechanism utilizes the control feedback path from the intelligent diagnosis unit 40 to the perturbation signal injection module 10.

[0144] The activation condition of this mechanism is that when the classification result of the current welding state by the defect evolution path tracking module is uncertain. This uncertain state is defined as the matching degree of the dynamic attributes (such as speed, curvature) of the evolution path with any one of the preset defect patterns (point, linear, periodic) being lower than a preset confidence threshold.

[0145] Under this condition, the intelligent diagnosis unit 40 automatically generates and sends a probing instruction to the perturbation signal injection module 10. The instruction contains the parameters of a specific probe signal sequence. Upon receiving the instruction, the perturbation signal injection module 10 suspends outputting the regular single-frequency sinusoidal probe signal and instead generates and injects the probe signal sequence specified in the instruction.

[0146] In one embodiment, the probe signal sequence is a linear sweep signal. The signal has a frequency f p (t) that linearly varies from a starting frequency f sweep to an ending frequency f start over a predetermined time period T end .

[0147]

[0148] In another embodiment, the probe signal sequence is a multi-tone signal, which is a linear superposition of sinusoidal waves of different frequencies.

[0149] During the targeted probing, the data acquisition unit 20 and the feature extraction unit 30 continuously operate. The feature extraction unit 30 employs a digital lock-in amplification algorithm to calculate the response of the system to the sweep signal, i.e., to obtain the response amplitude A p (f) and the response phase φ p (f) as functions of frequency.

[0150] The intelligent diagnosis unit 40 receives the frequency response features and compares them with a pre-set targeted-probing defect feature library. The library stores the frequency response curve features (e.g., resonance peaks or absorption valleys at specific frequency points) corresponding to known specific physical defects (e.g., internal cracks or inclusions of specific sizes).

[0151] If the real-time acquired frequency response curve matches an entry in the library, the intelligent diagnosis unit 40 confirms the initial uncertain diagnosis result as the specific defect type corresponding to the entry. If the response curve does not match any defect entry, the intelligent diagnosis unit 40 excludes the initial abnormality determination. After the process is completed, the intelligent diagnosis unit 40 sends an instruction to the perturbation signal injection module 10 to resume outputting the regular single-frequency probe signal.

[0152] In summary, the embodiment of the present application provides a high-frequency welded pipe production quality online monitoring intelligent diagnosis system and method. The system injects a low-energy, frequency-specific probe signal into the high-frequency welding circuit, and synchronously collects the high-frequency voltage and current waveforms and external physical field signals generated under the joint excitation of the main welding signal and the probe signal, realizing multi-dimensional information acquisition of the welding process. The feature extraction unit in the system extracts passive electromagnetic fingerprints and active response fingerprints from the responses to the main signal and the probe signal respectively through special algorithms, and fuses them into a unified and standardized welding state vector. The intelligent diagnosis unit uses a probability model based on historical health data to quickly detect abnormal states by calculating the Mahalanobis distance between the real-time state vector and the model. Further, the system classifies the dynamic mode of defects by analyzing the evolution path of the state vector in the feature space, and can trigger a closed-loop targeted probing mechanism to accurately identify specific defects by changing the form of the probe signal, thereby improving the accuracy and reliability of the diagnosis.

[0153] Referring to the drawings Figure 3 , Figure 3 is a flowchart of a high-frequency welded pipe production quality online monitoring intelligent diagnosis method according to an embodiment of the present application. The method can include the following steps in a specific embodiment:

[0154] S100, inject a probe signal of a predetermined frequency into the high-frequency welding circuit, and synchronously collect the high-frequency voltage waveform and the high-frequency current waveform generated by the high-frequency welding circuit under the joint excitation of the main welding frequency signal and the probe signal.

[0155] This step is performed by the perturbation signal injection module 10 and the data acquisition unit 20. The current and voltage signals are obtained by a Rogowski coil and a high-voltage differential probe respectively, and the multi-channel analog signals are converted into digital signal streams with high-precision time stamps by acquisition hardware based on the Precision Time Protocol (PTP) or central clock distribution system at a sampling rate of not less than 10 MS / s.

[0156] S200, based on the high-frequency voltage waveform and the high-frequency current waveform, extract at least one passive electromagnetic fingerprint and one active response fingerprint, and fuse to generate a welding state vector.

[0157] This step is performed by the feature extraction unit 30. First, the passive electromagnetic fingerprint is obtained by performing Hilbert transform on the voltage and current waveforms to calculate the instantaneous complex impedance, and by performing short-time Fourier transform on the current waveform to calculate the harmonic energy ratios. Meanwhile, the active response fingerprint is obtained by using a digital lock-in amplification algorithm to lock and extract the amplitude decay and phase shift of the probe signal response component from the voltage and / or current waveforms. Subsequently, all the extracted feature values are Z-score normalized and arranged in a predetermined order to form a multi-dimensional, dimensionless welding condition vector.

[0158] S300, compare the welding condition vector with a preset health model to determine the current high-frequency welded pipe welding quality.

[0159] This step is performed by the intelligent diagnosis unit 40. The degree of deviation of the current state from the normal working condition is quantitatively evaluated by calculating the Mahalanobis distance between the current welding condition vector and a mean vector and covariance matrix (i.e. health model) based on historical health data. The Mahalanobis distance is compared with a preset decision threshold, and if the distance exceeds the threshold, the current welding quality is determined to be abnormal.

[0160] In a further embodiment, after step S300, the method further comprises:

[0161] For the state vector sequence determined to be abnormal, the evolution path formed in the multi-dimensional feature space is tracked. By calculating the instantaneous speed and curvature of the path, the defect pattern is classified as one of point defect, linear defect or periodic defect, so as to preliminarily identify the nature of the defect.

[0162] In a still further embodiment, if the classification result of the aforementioned defect pattern is uncertain, a collaborative diagnosis refinement step is triggered:

[0163] The intelligent diagnosis unit 40 sends a control instruction to the perturbation signal injection module 10 to inject a specific probe signal sequence, such as a linear sweep signal. By analyzing the frequency response characteristics of the system to the specific sequence and comparing with a preset defect feature library, the type of defect is accurately confirmed.

[0164] In order to more clearly illustrate the technical solutions of the present application, a specific application embodiment is given below in conjunction with the drawings.

[0165] Embodiment:

[0166] On a production line for producing API 5L standard high-frequency straight seam welded pipes (ERW) with an outer diameter of 114 mm and a wall thickness of 4.0 mm, the main welding frequency f wThe frequency is 400 kHz, and the traveling speed of the steel pipe is 60 m / min. The online monitoring and intelligent diagnosis system of the present application is deployed on the production line.

[0167] Step 1, system initialization and normal monitoring:

[0168] After the system is started, the perturbation signal injection module 10 starts to work, and injects a sinusoidal probe signal with a frequency of f p = 50 kHz and a power of 5 watts into the induction coil of the high-frequency welding circuit. The frequency avoids the harmonics of the main welding frequency.

[0169] The Rogowski coil and high-voltage differential probe in the data acquisition unit 20 continuously collect high-frequency current and voltage signals, and digitize them at a sampling rate of 20 MS / s through a synchronous acquisition card. The feature extraction unit 30 calculates the welding state vector S(t) in real time, which is a five-dimensional vector in this embodiment, including the normalized equivalent resistance R eq , the equivalent reactance X eq , the second harmonic energy ratio HER2, the probe response amplitude decay A p , and the phase shift φ p .

[0170] The intelligent diagnosis unit 40 receives the vector sequence and calculates its Mahalanobis distance D M (t) from the preset health model (established based on 10,000 qualified samples of previous production).

[0171] Referring to the accompanying Figure 4 , Figure 4 , the Mahalanobis distance time series diagram displayed by the system monitoring interface in this embodiment. Under normal production conditions, the welding state is stable, and the calculated Mahalanobis distance D M (t) (shown by the solid line in the figure) fluctuates in a small range at a low level, and is always below the preset abnormality judgment threshold D threshold (shown by the dashed line in the figure).

[0172] Step 2, defect occurrence and abnormality detection:

[0173] At a certain time during production, a squeeze roller used for welding seam extrusion begins to produce slight and periodic radial jumping due to bearing wear. This mechanical vibration causes the convergence angle of the V-shaped area and the extrusion pressure to also change periodically at the same frequency. This is the root cause of a typical "periodic defect" that is difficult to detect by traditional methods.

[0174] This periodic physical change is immediately reflected in the electromagnetic fingerprint:

[0175] Passive fingerprint: periodic changes in the geometry of the V-shaped area cause the equivalent reactance X eqPeriodic fluctuations occur; changes in extrusion pressure affect contact resistance and molten pool condition, leading to an increase in equivalent resistance R. eq The harmonic energy ratio HER2 also oscillates accordingly.

[0176] Active fingerprinting: The periodic change in the overall impedance of the welding circuit also modulates the transmission path of the 50kHz probe signal, causing its response amplitude A to... p and phase φ p It also exhibits the same periodic fluctuations.

[0177] The synchronous periodic changes of these features cause the welding state vector S(t) to begin deviating from the healthy central region in the feature space. Therefore, the calculated Mahalanobis distance D... M (t) also began to rise and fall periodically. (See attached diagram) Figure 4 As shown in the "Defect Occurrence" area, D M The peak value of (t) repeatedly exceeded the anomaly detection threshold D. threshold The system immediately determined that the current welding status was abnormal.

[0178] Step 3: Defect Pattern Recognition and Diagnosis. Upon detecting an anomaly, the intelligent diagnostic unit 40 immediately activates the defect evolution path tracing module. This module analyzes the trajectory formed by the state vector S(t) sequence over a recent period (e.g., the last 2 seconds) in the five-dimensional feature space.

[0179] The defect evolution path tracing module, by calculating the velocity and curvature of the trajectory, discovered that it exhibits significant periodic characteristics. The module further analyzes any component of the state vector (e.g., X...) eq The frequency of the periodic fluctuation was calculated to be approximately 1.2 Hz by performing a Fast Fourier Transform on the time series (t).

[0180] Based on this, the system classifies the defect pattern as "periodic defect" and pushes a precise diagnostic message to the operator's monitoring interface: "Warning: Periodic welding defect detected. Defect frequency: approximately 1.2Hz. Possible cause: mechanical vibration. Please check the condition of the extrusion rollers and forming rollers."

[0181] Following the clear instructions, the operator quickly located the compression roller experiencing radial runout and replaced its worn bearing. After troubleshooting, see attached... Figure 4 As shown in the "Defect Elimination" region, the Mahalanobis distance D M (t) quickly falls below the threshold, and the system returns to normal monitoring status.

[0182] The application embodiment shows that the application can not only discover the abnormality of the welding quality in time through active-passive fingerprint fusion, but also classifies and traces the root mode of the defect through in-depth analysis on the defect evolution path, thereby greatly improving the fault troubleshooting efficiency and accuracy and guaranteeing the production quality of the high-frequency welded pipe.

[0183] Although embodiments of the application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-frequency welded pipe production quality online monitoring and intelligent diagnostic system, characterized in that, include: The data acquisition unit is configured to synchronously acquire high-frequency voltage waveforms and high-frequency current waveforms in the high-frequency welding circuit. The high-frequency voltage waveforms and high-frequency current waveforms are generated by the high-frequency welding circuit under the joint excitation of the main welding frequency signal and the injected probe signal. The feature extraction unit, connected to the data acquisition unit, is configured to extract at least one passive electromagnetic fingerprint and one active response fingerprint based on the high-frequency voltage waveform and the high-frequency current waveform, and fuse them to generate a welding state vector; the passive electromagnetic fingerprint includes at least one of instantaneous complex impedance and / or harmonic energy ratio, and the active response fingerprint is the response feature of the high-frequency welding circuit to the probe signal; The intelligent diagnostic unit, connected to the feature extraction unit, is configured to compare the welding state vector with a preset health model to determine the current welding quality of the high-frequency welded pipe.

2. The intelligent diagnostic system for online monitoring of high-frequency welded pipe production quality according to claim 1, characterized in that, The system also includes: A perturbation signal injection module, coupled to the high-frequency welding circuit, is configured to inject the probe signal into the high-frequency welding circuit; the frequency of the probe signal is different from the fundamental frequency and integer multiples of the harmonic frequency of the main welding frequency signal.

3. The intelligent diagnostic system for online monitoring of high-frequency welded pipe production quality according to claim 1, characterized in that, The feature extraction unit extracts the passive electromagnetic fingerprint, specifically configured as follows: The ratio between the high-frequency voltage waveform and the high-frequency current waveform is calculated to obtain the instantaneous complex impedance, which includes the equivalent resistance characterizing energy dissipation and the equivalent reactance characterizing electromagnetic energy storage. and / or The high-frequency current waveform is subjected to spectral analysis to obtain the harmonic energy ratio, which characterizes the degree of nonlinearity in the welding process.

4. The intelligent diagnostic system for online monitoring of high-frequency welded pipe production quality according to claim 1, characterized in that, The feature extraction unit extracts the active response fingerprint, specifically configured as follows: A digital lock-in amplification algorithm is used to lock and extract the response component caused by the probe signal from the high-frequency current waveform and / or high-frequency voltage waveform to obtain the active response fingerprint; The active response fingerprint includes amplitude attenuation and / or phase shift of the response component.

5. The online monitoring and intelligent diagnostic system for high-frequency welded pipe production quality according to claim 2, characterized in that, The intelligent diagnostic unit compares the welding state vector with a preset health model, specifically configured as follows: The health model is a probability distribution model established based on multiple historical welding state vectors collected during the production of qualified welded pipes. The intelligent diagnostic unit quantifies the degree of deviation by calculating the Mahalanobis distance between the current welding state vector and the health model, and determines the welding quality based on the degree of deviation.

6. The intelligent diagnostic system for online monitoring of high-frequency welded pipe production quality according to claim 5, characterized in that, The intelligent diagnostic unit also includes: The defect evolution path tracking module is configured to store a time-series welding state vector to form an evolution path and to identify defect patterns based on the dynamic properties of the evolution path.

7. The intelligent diagnostic system for online monitoring of high-frequency welded pipe production quality according to claim 6, characterized in that, The defect evolution path tracking module identifies defect patterns based on the dynamic attributes of the evolution path, and is specifically configured as follows: The geometric and dynamic properties of the evolution path in the multidimensional feature space are analyzed, including the velocity, curvature and direction changes of the trajectory, to classify the defect pattern into one of point defects, linear defects or periodic defects.

8. The intelligent diagnostic system for online monitoring of high-frequency welded pipe production quality according to claim 6, characterized in that, The system is also configured to: When the defect evolution path tracking module identifies a preset defect pattern, it triggers the perturbation signal injection module to adjust or inject a specific probe signal sequence to target and confirm the defect pattern.

9. The intelligent diagnostic system for online monitoring of high-frequency welded pipe production quality according to claim 1, characterized in that, The data acquisition unit is also configured to synchronously acquire at least one external physical signal, which includes at least one of the thermal field information of the weld area or the geometric information of the weld bead. The feature extraction unit is also configured to integrate the features extracted from the external physical signal into the welding state vector.

10. A method for online monitoring and intelligent diagnosis of high-frequency welded pipe production quality, based on the system described in any one of claims 1-9, characterized in that, Includes the following steps: The high-frequency voltage and current waveforms generated by the high-frequency welding circuit under the joint excitation of the main welding frequency signal and the injected probe signal are acquired synchronously. Based on the high-frequency voltage waveform and high-frequency current waveform, at least one passive electromagnetic fingerprint and one active response fingerprint are extracted and fused to generate a welding state vector; the passive electromagnetic fingerprint includes at least one of instantaneous complex impedance and / or harmonic energy ratio, and the active response fingerprint is the response characteristics of the high-frequency welding circuit to the probe signal; The welding state vector is compared with a preset health model to determine the current welding quality of the high-frequency welded pipe.

Images