Intermediate frequency resistance welding transformer cooling water path structure

Abstract

The present application relates to the technical field of welding equipment, and particularly relates to a cooling water path structure of a medium-frequency resistance welding transformer. The structure comprises a transformer body, the transformer body is provided with a protective shell, a water path plate and a transformer heat core assembly; mirror image symmetrical fin mounting structures are integrally formed on the left and right sides of the water path plate, the symmetrical fin mounting structures are wing-shaped mounting bosses horizontally extending outward from the side surface of the water path plate, and mounting holes are formed in the plate body of the wing-shaped mounting bosses; a Chinese character type cooling water path for circulating cooling water is formed in the water path plate through deep hole processing, and the Chinese character type cooling water path is formed by a longitudinal main water path extending along the vertical central axis of the water path plate and two groups of transverse branch paths perpendicularly connected to the longitudinal main water path and arranged in parallel in the upper and lower positions. The present application eliminates the interference contradiction between the cooling flow channel and the mounting hole, improves the assembly efficiency and the structural stability, and solves the problems of uneven heat dissipation and local overheating of the existing flow channel.

Description

Technical Field

[0001] This invention relates to the field of welding equipment technology, and in particular to a cooling water circuit structure for a medium-frequency resistance welding transformer. Background Technology

[0002] As the core power component of welding machines, medium-frequency resistance welding transformers are widely used in the field of metal welding and processing. During their operation, the internal core components of the transformer generate a large amount of heat due to Joule heat and eddy current losses in the iron core and windings. If the heat cannot be dissipated in a timely and efficient manner, the internal temperature of the transformer will rise sharply, which will affect the insulation performance of the windings and the magnetic permeability of the iron core, reduce the working stability and service life of the transformer, and even cause equipment failures such as winding burnout and iron core deformation. Therefore, forced water cooling has become an indispensable heat dissipation method for medium-frequency resistance welding transformers. Currently, the cooling water channels of existing medium-frequency resistance welding transformers mostly adopt simple straight, U-shaped, or annular flow channel designs. The cooling water channels are independently arranged from the transformer mounting structure, resulting in a cumbersome overall layout, large space occupation, and potential structural interference between the water channels and mounting holes during assembly, affecting assembly efficiency and structural stability. Furthermore, existing cooling water channels lack targeted flow channel optimization for the transformer's core heat-generating components, leading to uneven distribution of cooling water within the channels, poor heat dissipation in areas of concentrated heat generation, and a tendency for localized overheating. Moreover, the flow channel connection nodes often use right-angle transition structures, resulting in high flow resistance and water stagnation, further reducing heat dissipation uniformity. Regarding the transformer mounting structure, existing technologies often employ independent mounting brackets separated from the water channel plate. This easily leads to stress concentration during installation and, under long-term exposure to welding vibrations, cracks can occur at the connection points between the mounting brackets and the water channel plate, compromising the cooling water channel's seal and causing cooling water leakage. Furthermore, existing methods for detecting leaks in the cooling water circuit of medium-frequency resistance welding transformers mostly employ offline static detection, which cannot monitor leaks in real time under the dynamic operating conditions of transformer welding under load. This is especially true for micro-crack leaks in critical areas such as the arc-shaped cross-pipe section of the lower transverse shunt branch, which are difficult to accurately identify and locate. Moreover, existing detection methods do not take into account the flow channel topology characteristics of the cooling water circuit and the transformer's operating conditions, resulting in low detection accuracy and weak anti-interference capabilities. They cannot detect micro-leaks and locate their positions in a timely manner. Once a leak occurs, it will not only lead to a decrease in cooling efficiency but may also damage the transformer's internal core components due to cooling water leakage, causing equipment failure and safety hazards. Summary of the Invention

[0003] This invention overcomes the shortcomings of the prior art and provides a cooling water circuit structure for a medium-frequency resistance welding transformer.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention discloses a cooling water circuit structure for a medium-frequency resistance welding transformer, including a transformer body; the transformer body includes an external protective shell, a water circuit plate embedded inside the protective shell, and a transformer heating core component; the water circuit plate is a rigid metal plate structure, and symmetrical wing-shaped mounting structures are integrally formed on the left and right sides of the water circuit plate, which are horizontally extending wing-shaped mounting bosses from the side of the water circuit plate. The wing-shaped mounting bosses have mounting holes that penetrate along the thickness direction for locking and fixing with a mounting base; the water circuit plate... The interior is formed by deep hole machining to create a U-shaped cooling water channel for the flow of cooling water. The U-shaped cooling water channel consists of a longitudinal main water channel extending along the vertical central axis of the water channel plate and two sets of transverse branch channels arranged vertically and parallel to the longitudinal main water channel. The longitudinal main water channel and the two sets of transverse branch channels intersect to form a closed-loop U-shaped circulating cooling channel. The lower transverse branch channel of the U-shaped cooling water channel is provided with an arc-shaped cross-connection section. The cross-connection section transversely crosses the mounting hole area on the symmetrical wing mounting structure, so that the cooling water channel and the transformer mounting and fixing structure are integrated into one unit.

[0005] This invention also discloses a leakage detection method for the cooling water circuit structure of a medium-frequency resistance welding transformer, applicable to the cooling water circuit structure of any of the aforementioned medium-frequency resistance welding transformers, comprising the following steps: S1. When the medium frequency resistance welding transformer is under load welding condition and the cooling water circuit is circulating, the flow time sequence signal of the inlet end and the outlet end of the upper and lower transverse branch is collected simultaneously to construct a multi-dimensional flow observation matrix that reflects the topological constraint relationship of the water circuit. S2. Based on the multidimensional flow observation matrix, calculate the normalized leakage flow difference factor to characterize the leakage degree of the lower transverse branch; S3. Compare the trend deviation of the normalized leakage flow difference factor with the adaptive leakage threshold cloud map based on transformer operating condition identification to generate the evolution trajectory of the difference factor to describe the dynamic diffusion behavior of leakage. S4. Perform modal decomposition on the evolution trajectory of the difference factor to extract the non-stationary leakage pulse characteristic modal components induced by the opening and closing of microcracks in the cross-pipe section. S5. When the intermittent energy entropy of the non-stationary leakage pulse characteristic mode component exceeds the preset leakage confidence interval, it is determined that a leakage has occurred in the cooling water circuit and the leakage location is marked in the cross-connection section area.

[0006] This invention addresses the technical deficiencies in the prior art and has the following beneficial effects: By integrating the design of the Chinese character-shaped cooling water channel and the symmetric wing installation structure, the interference contradiction between the cooling flow channel and the installation hole is eliminated at the structural level, while simplifying the overall layout, improving the assembly efficiency and structural stability; the closed-loop flow channel composed of the longitudinal main water channel and the upper and lower groups of transverse shunt branches and combined with the arc smooth transition nodes effectively reduces the flow resistance, avoids water flow retention, and realizes targeted and uniform heat dissipation for the lower region where the transformer's heat-generating core components are located, solving the problems of uneven heat dissipation and local overheating of the existing flow channels. The supporting leakage detection method makes full use of the unique flow resistance topological constraint relationship of the Chinese character-shaped water channel. Through the multi-dimensional matrix observation of the synchronous flow signal of the welding cycle and the Kirchhoff node constraint calculation, the weak flow anomaly caused by the micro-leakage is transformed into a normalized difference factor, and then combined with the adaptive chirp mode decomposition and the intermittent energy entropy criterion, the real-time online detection and accurate positioning of the opening and closing type leakage of the micro-cracks in the cross-connecting pipe section are realized under the dynamic working condition of the transformer with load welding, overcoming the defect that the traditional offline static detection cannot identify the intermittent micro-leakage, and improving the sensitivity, accuracy and working condition adaptability of the leakage detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0008] Figure 1 It is a schematic diagram of the overall structure of the transformer body of the present invention; Figure 2 It is a schematic diagram of the first internal structure of the transformer body of the present invention; Figure 3 It is a schematic diagram of the second internal structure of the transformer body of the present invention; Figure 4 It is a schematic diagram of the cross-section of the water channel plate in the transformer body of the present invention; In the figure: 101, transformer body; 102, protective housing; 103, water channel plate; 104, transformer heat-generating core component; 105, wing-shaped mounting boss; 106, mounting hole; 108, longitudinal main water channel; 109, transverse shunt branch; 201, cross-connecting pipe section; 202, heat dissipation plate. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0009] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the following will further describe the present invention in detail with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0010] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0011] like Figure 1 , 2 As shown in Figure 3, the present invention discloses a cooling water circuit structure for a medium-frequency resistance welding transformer, including a transformer body 101; the transformer body 101 includes an external protective shell 102, a water circuit plate 103 embedded inside the protective shell 102, and a transformer heating core component 104; the water circuit plate 103 is a rigid metal plate structure, and the left and right sides of the water circuit plate 103 are integrally formed with symmetrical wing mounting structures distributed in a mirror image. The symmetrical wing mounting structures are wing-shaped mounting bosses 105 extending horizontally outward from the side of the water circuit plate 103. The wing-shaped mounting bosses 105 have mounting holes 106 that are provided through the plate along the thickness direction for locking and fixing with the mounting base.

[0012] like Figure 4 As shown, the interior of the water circuit plate 103 is formed by deep hole machining to create a U-shaped cooling water circuit for the flow of cooling water. The U-shaped cooling water circuit consists of a longitudinal main water circuit 108 extending along the vertical central axis of the water circuit plate 103 and two sets of transverse branch circuits 109 arranged vertically and parallel to the longitudinal main water circuit 108. The longitudinal main water circuit 108 and the two sets of transverse branch circuits 109 intersect to form a closed-loop U-shaped circulating cooling channel. The lower transverse branch circuit 109 of the U-shaped cooling water circuit is provided with an arc-shaped cross-connection section 201. The cross-connection section 201 transversely crosses the mounting hole 106 area on the symmetrical wing mounting structure, so that the cooling water circuit and the transformer mounting and fixing structure are integrated. The side wall of the protective shell 102 is provided with water connector assembly holes corresponding to the inlet and outlet positions of the U-shaped cooling water circuit, for connecting to external cooling water supply pipes.

[0013] The connection points between the longitudinal main water channel 108 and the two sets of transverse branch channels 109 adopt a smooth arc transition structure. The two wing-shaped mounting bosses 105 of the symmetrical wing mounting structure are strictly symmetrically distributed along the vertical central axis of the water channel plate 103, and the thickness of the wing-shaped mounting bosses 105 is consistent with the thickness of the main body of the water channel plate 103. The inner wall of the cross-connection section 201 is a smooth curved surface, and a safety clearance is reserved between the cross-connection section 201 and the mounting hole 106 to avoid structural interference between the water channel and the mounting hole 106, while ensuring smooth flow of cooling water within the cross-connection section 201.

[0014] It should be noted that the working process of the cooling water circuit structure of the medium frequency resistance welding transformer of the present invention is as follows: First, the entire cooling water circuit structure is locked and fixed to the assembly base by the symmetrical wing-shaped mounting structure integrally formed on the left and right sides of the water circuit plate 103. The wing-shaped mounting bosses 105 are strictly symmetrically distributed along the vertical central axis of the water circuit plate 103, and their plate thickness is consistent with the thickness of the main body of the water circuit plate 103, which can effectively avoid stress concentration during installation. At the same time, a safety clearance is reserved between the mounting hole 106 on the wing-shaped mounting bosses 105 and the arc-shaped cross-connection section 201, which completely avoids structural interference between the water circuit and the mounting hole 106, realizing the integration of the cooling water circuit and the mounting and fixing structure. The system is assembled in an integrated manner. After assembly, the external cooling water supply pipes are connected to the water inlet assembly holes on the side wall of the protective shell 102, allowing cooling water to smoothly enter the T-shaped cooling water channel inside the water channel plate 103, which is machined through deep holes. After entering from the inlet, the cooling water first flows into the longitudinal main water channel 108 extending along the vertical central axis of the water channel plate 103. As it flows downward along the longitudinal main water channel 108, it gradually branches into two sets of horizontal branch channels 109 arranged in parallel. Because the connection nodes between the longitudinal main water channel 108 and the two sets of horizontal branch channels 109 adopt a smooth arc transition structure, the flow resistance can be effectively reduced, water stagnation can be avoided, and the smooth distribution of cooling water can be ensured. The upper transverse branch 109 dissipates heat in the upper area of ​​the water circuit board 103, while the lower transverse branch 109 extends laterally through the arc-shaped crossover pipe section 201. This crossover pipe section 201 laterally crosses the mounting hole 106 area on the symmetrical wing mounting structure, and its inner wall is a smooth curved surface, further improving the smoothness of cooling water flow. At the same time, the lower transverse branch 109 is connected to the built-in water channel of the heat sink 202 at the transformer heat-generating core component 104. After the cooling water flows through the arc-shaped crossover pipe section 201, it enters the built-in water channel of the heat sink 202, directly absorbing the large amount of heat generated by the transformer heat-generating core component 104, and achieving targeted heat dissipation in the lower heat-concentrated area. After absorbing heat, the cooling water flows back along the internal water channel of the heat dissipation plate 202 to the lower horizontal branch channel 109, and then converges into the longitudinal main water channel 108, finally flowing out from the outlet end, completing a closed-loop circulation heat dissipation. This continuous circulation can achieve uniform heat dissipation over the entire transformer body 101, ensuring that the temperature of the transformer's core heat-generating component 104 is maintained within a reasonable range. Throughout the entire operation, the protective shell 102 provides good protection for the internal water channel plate 103, the transformer's core heat-generating component 104, and the cooling water channel. The assembly gap between the shell and the water channel plate 103 is sealed to effectively prevent cooling water leakage, ensuring the sealing performance and operational stability of the cooling water channel.

[0015] In summary, the T-shaped cooling water circuit of this invention, through the closed-loop design of the longitudinal main water circuit 108 and the upper and lower sets of transverse branch circuits 109, combined with the connection with the built-in water circuit of the heat sink 202, can achieve uniform heat dissipation over the entire transformer body 101, especially for targeted heat dissipation of the transformer's core heat-generating component 104 located at the bottom. This effectively solves the problems of uneven heat dissipation in the middle cooling water circuit and poor heat dissipation in the lower concentrated heat-generating area. At the same time, the smooth transition of the connecting nodes and the smooth inner wall design of the arc-shaped cross-connection section 201 significantly reduces water flow resistance, avoids water stagnation, and further improves heat dissipation efficiency. The symmetrical wing mounting structure is integrally formed with the water circuit plate 103, realizing the integration and fixing of the cooling water circuit. The integrated structure not only simplifies the overall layout and reduces space occupation, but also avoids stress concentration problems caused by the separate design of independent mounting brackets and water circuit board 103, improving assembly efficiency and structural stability. At the same time, the reserved safety clearance completely avoids structural interference between the water circuit and the mounting hole 106. In addition, the combination of integrated structure, sealing protection design and smooth curved surface design of arc-shaped cross-connection section 201 effectively improves the sealing performance of the cooling water circuit and reduces the risk of cooling water leakage. The overall structural design is reasonable, the assembly is convenient, the heat dissipation efficiency is high, and the structural stability is strong, which can effectively extend the service life of medium frequency resistance welding transformers and meet the requirements of efficient, stable and long-term operation.

[0016] This invention also discloses a leakage detection method for the cooling water circuit structure of a medium-frequency resistance welding transformer, applicable to the cooling water circuit structure of any of the aforementioned medium-frequency resistance welding transformers, comprising the following steps: S1. When the medium frequency resistance welding transformer is under load welding condition and the cooling water circuit is circulating, the flow time sequence signal of the inlet end and the outlet end of the upper and lower transverse branch is collected simultaneously to construct a multi-dimensional flow observation matrix that reflects the topological constraint relationship of the water circuit. S2. Based on the multidimensional flow observation matrix, calculate the normalized leakage flow difference factor to characterize the leakage degree of the lower transverse branch; S3. Compare the trend deviation of the normalized leakage flow difference factor with the adaptive leakage threshold cloud map based on transformer operating condition identification to generate the evolution trajectory of the difference factor to describe the dynamic diffusion behavior of leakage. S4. Perform modal decomposition on the evolution trajectory of the difference factor to extract the non-stationary leakage pulse characteristic modal components induced by the opening and closing of microcracks in the cross-pipe section. S5. When the intermittent energy entropy of the non-stationary leakage pulse characteristic mode component exceeds the preset leakage confidence interval, it is determined that a leakage has occurred in the cooling water circuit and the leakage location is marked in the cross-connection section area.

[0017] Specifically, S1 is: At the starting moment of synchronously triggering the welding pulse of the medium-frequency resistance welding transformer, equal-duration segmented window truncation is performed on the original flow time-series signals output by the flow sensor at the water inlet end, the flow sensor at the water outlet end of the upper horizontal shunt branch, and the flow sensor at the water outlet end of the lower horizontal shunt branch, to obtain the water inlet end flow window sequence, the upper branch flow window sequence, and the lower branch flow window sequence that are synchronously locked with a single welding cycle; For the water inlet end flow window sequence, the upper branch flow window sequence, and the lower branch flow window sequence, the flow mean value feature and the flow fluctuation kurtosis feature within the window are respectively extracted, and the flow mean value feature and the flow fluctuation kurtosis feature of each window are feature-stitched according to the time sequence index, to form the water inlet end feature flow, the upper branch feature flow, and the lower branch feature flow that characterize the flow states of each measurement point; According to the flow resistance topological constraint relationship at the cross-connection pipe section node of the Chinese character-shaped cooling water path, the water inlet end feature flow, the upper branch feature flow, and the lower branch feature flow are substituted into the matrix expression framework based on the fluid mechanics continuity theorem point by point along the time axis, and the window flow imbalance vector reflecting the degree of flow distribution imbalance of each branch within the window is calculated window by window; The window flow imbalance vectors of consecutive multiple welding cycles are longitudinally stacked in chronological order, to construct an imbalance time-series stack matrix with the welding cycle number as the row index and the branch node flow deviation as the column vector; Element-by-element differential operation is performed on the imbalance time-series stack matrix and the flow resistance topological constraint matrix under the normal state of the pre-calibrated water path, to obtain a multi-dimensional flow observation matrix reflecting the degree of deviation of the flow distribution of the Chinese character-shaped water path from the topological constraint under the load condition.

[0018] It should be noted that the flow resistance topological constraint matrix is a set of reference quantities obtained through pre-calibration, which reflects the natural formed flow distribution inherent proportional relationship between the longitudinal main water path and the upper and lower horizontal shunt branches in the Chinese character-shaped cooling water path under the normal state of the water path being intact and without any leakage. During calibration, the cooling water can be circulated under the healthy condition of the water path without leakage, and the flow signals at the water inlet end and the water outlet ends of the upper and lower branches are synchronously recorded, and the flow distribution coefficients within each window are extracted, and a matrix with the welding cycle as the row and the flow distribution coefficients of each branch flow node as the elements is constructed, that is, a nominal matrix representing the normal topology constraint of the water path is formed.

[0019] During the monitoring process, for the obtained inlet-end characteristic flow, upper-branch characteristic flow, and lower-branch characteristic flow, substituting them point by point along the time axis can be understood as, within each welding cycle window, regarding the characteristic values of the three-way characteristic flows in the current window as the measured variables of the three nodes in the flow channel network and substituting them into the matrix expression framework constructed based on the fluid mechanics continuity theorem. The construction logic of this framework is as follows: For the middle-shaped waterway, there are clear flow channel branch and confluence relationships at the cross-connecting pipe segment nodes. The inflow at any one end should be equal to the sum of the outflows at the other two ends, which is a direct manifestation of Kirchhoff's flow conservation in the specific waterway topology. During actual operation, taking a certain window as an example, use the inlet-end characteristic value, upper-branch characteristic value, and lower-branch characteristic value of this window as the three elements of a vector, and generate a deviation characterization quantity through a linear combination method. For example, a distribution reference system can be set. Under this reference system, the inlet-end characteristic value is mapped as the expected flow rate benchmark for the branch, and the deviations between the upper and lower branch characteristic values and their respective expected flow rates are arranged as the deviation components of this window. This process does not introduce additional assumptions, but only extracts the degree of imbalance in the flow rate distribution of each branch within one observation as an ordered numerical vector through matrix arrangement. After performing this operation for each window in sequence, a window flow imbalance vector reflecting the degree of imbalance in the branch flow rate distribution within each window is obtained.

[0020] Vertically stack the window flow imbalance vectors into an imbalance time series stack matrix, and perform a对位差分 (对位差分 is not a standard English term, perhaps it should be "pairwise difference") with the flow resistance topology constraint matrix. Its actual meaning is to perform a quantitative comparison, window by window and node by node, between the real-time extracted flow imbalance state and the pre-calibrated normal topology distribution state. The result of the difference is the multi-dimensional flow observation matrix. Each element of it intuitively indicates the deviation amplitude and direction of the actual flow rate distribution relative to the normal topology constraint at a certain welding cycle and a certain waterway branch node, making the tiny leakage in the lower branch able to be highlighted in the form of an abnormal increase in the continuity of the corresponding column element values in the observation matrix.

[0021] The specific content of S2 is as follows: Extract the flow distribution deviation vectors corresponding to each window row by row according to the welding cycle index from the multi-dimensional flow observation matrix, and form a deviation vector sequence reflecting the flow rate deviation amplitudes of each branch in the middle-shaped waterway within a single welding cycle; Based on the flow resistance topology connection relationship at the cross-connecting pipe segment nodes of the middle-shaped cooling waterway, construct a Kirchhoff node association constraint model with the inlet-end confluence node, upper-branch shunt node, and lower-branch cross-connecting pipe segment shunt node as topological units, and substitute each deviation vector in the deviation vector sequence into the Kirchhoff node association constraint model in sequence to solve and obtain a node flow constraint residual vector representing the flow conservation residuals of each node; It should be noted that the Kirchhoff node association constraint mode converts the physical connection relationship of the Chinese character-shaped cooling water path at the cross-connection pipe segment nodes into a set of node flow conservation conditions. Specifically, the water inlet convergence node, the upper branch diversion node, and the lower branch cross-connection pipe segment diversion node can be regarded as three topological units in the flow channel network respectively. According to the inflow-outflow connection situation of each node in the actual water path, the flow balance constraint relationship that each node should satisfy is established. Taking the cross-connection pipe segment diversion node as an example, in the normal non-leakage state, the inflow from the longitudinal main water path direction should be equal to the sum of the outflows to the built-in water path of the heat dissipation plate and the lower branch water outlet direction; when there is a microcrack leakage, a part of the leakage loss that does not enter the downstream sensor will be included in the actual outflow of this node, resulting in the measured flow relationship of this node deviating from the conservation condition. During the solution, the deviation vectors corresponding to each window in the deviation vector sequence are regarded as the measured deviation values of each node under the current window, and are substituted into this set of node conservation constraint relationships for verification, and then the additional compensation amounts required for each node to satisfy the conservation condition under the current window can be solved. The compensation amount is the node flow constraint residual, which is arranged in the order of nodes to form a node flow constraint residual vector, which characterizes the degree of deviation of each node from the ideal flow conservation within the current window. Among them, the residual component of the lower branch node is particularly sensitive to leakage.

[0022] Arrange the node flow constraint residual vectors corresponding to consecutive welding cycles in chronological order to generate a node flow constraint residual matrix, and impose local flow resistance sensitivity screening on the node flow constraint residual matrix according to the branch correspondence relationship between the lower horizontal diversion branch and the cross-connection pipe segment in the flow resistance topology, and separate the lower branch leakage residual contribution component matrix that is only related to the leakage source strength of the lower horizontal diversion branch from it; It should be noted that after obtaining the node flow constraint residual matrix of consecutive welding cycles, it is necessary to separate the contribution components that are only related to the leakage source strength of the lower horizontal diversion branch. The reason is that there are differences in the sensitivity of different branches in the Chinese character-shaped cooling water path to changes in flow resistance. Since there is no influence of the cross-pipe structure in the upper branch, the node residuals of the upper branch are mainly caused by the overall fluctuation of the working conditions, showing a slow and same-direction change between each window; while the lower branch contains an arc-shaped cross-connection pipe segment, and the flow resistance characteristics of this pipe segment will undergo instantaneous disturbances due to the opening and closing of microcracks, resulting in mutation components different from other nodes in the node residuals of the lower branch. During the screening, according to the branch correspondence relationship between the lower horizontal diversion branch and the cross-connection pipe segment in the flow resistance topology, the mode classification of each node residual component in the residual matrix along the time direction can be carried out, and the subset of residual components corresponding to the lower branch nodes and conforming to the flow resistance mutation characteristics of the cross-connection pipe segment can be extracted, and recombined into the lower branch leakage residual contribution component matrix. This operation is equivalent to performing a spatial filtering in the residual space to suppress the common-mode residual components that are not related to leakage.

[0023] The singular feature direction of the leakage mode is extracted from the leakage residual contribution component matrix of the lower branch to obtain the main leakage feature direction vector that can best characterize the leakage amplitude fluctuation of the lower transverse branch. The main leakage characteristic direction vector is projected onto the direction of the average total flow rate at the inlet, and normalized and scaled using the continuous constraint of the overall water flow, and finally the normalized leakage flow rate difference factor is calculated.

[0024] It should be noted that the residual contribution component matrix of the lower branch leakage contains residual components related to leakage under multiple welding cycle windows, but these components may be distributed along different directions of change. By extracting singular feature directions, a feature direction that best represents the main trend of leakage amplitude fluctuation can be identified from the matrix. The physical meaning is that although the leakage residual of each window manifests differently at different nodes, the residual vectors of all windows are often most discretely distributed along a certain dominant direction in high-dimensional space. This direction captures the most important pattern of abnormal flow changes caused by leakage. For example, as the leakage gradually intensifies, the residual vector of the lower branch node shows an approximately linear growth trend in the inflow-outflow direction of the cross-pipe section. This growth direction can be extracted as the main leakage feature direction vector. Essentially, the vector is an optimal compressed representation of leakage information in the lower branch residual space. Its direction indicates the basic distribution pattern of leakage impact among nodes, and its length reflects the relative strength of this pattern.

[0025] The main leakage characteristic direction vector is projected onto the average total flow rate at the inlet and then normalized and scaled. This aims to eliminate the impact of variations in total flow rate caused by load current changes on the leakage index. Since the average total flow rate at the inlet varies under different welding current settings, the absolute deviation caused by leakage will also change accordingly. Directly using the vector magnitude as the leakage index would lead to condition dependence. By projecting the vector onto the average total flow rate, the component of the leakage deviation that is proportional to the total flow rate is extracted. Then, normalization and scaling are performed based on the continuous constraint of the overall water flow, a dimensionless normalized leakage flow rate difference factor is obtained. An increase in the factor value directly corresponds to an increase in the degree of leakage in the lower transverse branch, and it exhibits good adaptability to changes in operating conditions.

[0026] Specifically, S3 is: The welding current waveform signal and the inlet water temperature signal of the T-shaped cooling water circuit are synchronously acquired during the welding process of the transformer under load. The effective value of the welding current and the inlet water temperature value are used to form a binary group of the current operating conditions. Based on the pre-established adaptive leakage threshold cloud map, the upper limit and lower limit of the adaptive leakage threshold of the operating conditions that match the binary group of the current operating conditions are found and extracted to form the adaptive leakage threshold range under the current operating conditions. It should be noted that the adaptive leakage threshold cloud map establishment process is as follows: Under a healthy state where the water circuit is confirmed to be leak-free, the transformer is subjected to various combinations of welding current and inlet water temperature that may occur during actual use. Steady-state fluctuation data of the normalized leakage flow difference factor are continuously collected under each combination of operating conditions. The upper and lower bounds of the normal fluctuation of this factor under each operating condition are statistically derived. All operating points and their corresponding upper and lower bound values ​​are used to construct a threshold distribution map with the effective value of welding current and the inlet water temperature as two-dimensional coordinates, which is the adaptive leakage threshold cloud map. Essentially, the cloud map discretizes and stores the concept of the normal random fluctuation range of the difference factor under leak-free operating conditions using operating condition pairs as indexes. This allows for the direct retrieval of the matching threshold range based on the current welding current and inlet water temperature during subsequent actual monitoring, eliminating the need to use a uniform fixed threshold applicable to all operating conditions when determining leakage.

[0027] The time-series values ​​of the normalized leakage flow difference factor on multiple consecutive welding cycles are compared with the membership degree of the adaptive leakage threshold interval at the corresponding cycle time. It is determined whether the normalized leakage flow difference factor of each cycle falls within or outside the adaptive leakage threshold interval. The deviation direction and deviation magnitude value of the difference factor of each cycle relative to the center of the adaptive leakage threshold interval are extracted to form a leakage deviation state sequence. The deviation amplitude values ​​of each cycle in the leakage deviation state sequence are subjected to sliding weighted cumulative deviation trend enhancement processing along the time axis. The increment of the same-direction deviation amplitude between adjacent cycles is accumulated as the trend cumulative deviation, while suppressing the instantaneous reverse deviation caused by measurement noise, and generating a trend cumulative deviation sequence that reflects the continuous change trend of the leakage deviation degree. The trend cumulative deviation sequence is mapped to a two-dimensional deviation phase plane with the welding cycle number as the horizontal axis and the trend cumulative deviation as the vertical axis. The points on the two-dimensional deviation phase plane are connected in time evolution order to form an initial deviation phase trajectory describing the dynamic diffusion behavior of the normalized leakage flow difference factor relative to the adaptive leakage threshold range. The initial deviation phase trajectory is smoothed and the trend direction feature is extracted based on phase space reconstruction. The high-frequency reciprocating fluctuation component caused by instantaneous disturbance of the operating condition is removed from the initial deviation phase trajectory, and the dominant trend skeleton that can reflect the consistency of leakage diffusion direction and the change law of diffusion rate is retained to generate the difference factor evolution trajectory.

[0028] It should be noted that during the generation of the evolution trajectory, the threshold range for adaptive operation is retrieved using cloud maps. The values ​​of the difference factor at each welding cycle are compared with the upper and lower bounds of the threshold at the corresponding time point to obtain the deviation direction and amplitude for each cycle, forming a leakage deviation state sequence. Although this sequence reflects whether the difference factor of each cycle exceeds the normal range, due to measurement noise and instantaneous operating condition disturbances, occasional reverse deviations may occur in individual cycles. Therefore, subsequent cumulative deviation trend enhancement processing using sliding weighting continuously accumulates the amplitude of the same-direction deviation between adjacent cycles, while suppressing isolated instantaneous reverse deviations that are inconsistent with the deviation direction of the preceding and following cycles, thereby obtaining a trend cumulative deviation sequence. This sequence reflects the continuous change of the leakage deviation degree on the time axis, rather than an isolated point-by-point state.

[0029] The cumulative deviation sequence is mapped onto a two-dimensional deviation phase plane and connected in chronological order to form the initial deviation phase trajectory. This trajectory visually represents the overall movement of the difference factor relative to the threshold range in the phase plane. However, due to the complex electromagnetic environment during welding, the trajectory inevitably contains high-frequency reciprocating fluctuations caused by instantaneous disturbances. Finally, by applying smoothing and trend direction feature extraction processing based on phase space reconstruction to the initial trajectory, high-frequency irregular disturbances are eliminated, and the dominant skeleton reflecting the consistency of leakage diffusion direction and the variation law of diffusion rate is retained in the trajectory, thereby generating the difference factor evolution trajectory. Its morphological features such as direction, slope, and curvature are directly related to the diffusion dynamics of leakage in the lower transverse branch.

[0030] Specifically, S4 is: The time-frequency energy distribution spectrum of the evolution trajectory of the difference factor is constructed along the welding cycle time axis, transforming the evolution trajectory of the difference factor from a single amplitude-time domain to a three-dimensional joint characterization space of time-frequency-energy, thereby obtaining the time-frequency energy distribution spectrum. It should be noted that the evolution trajectory of the difference factor can be segmented along the time axis corresponding to the welding cycle number by applying a sliding time window. Time-frequency transformation is performed on the signal segment within each time window to obtain the instantaneous spectrum corresponding to the center moment of that time window. Arranging the instantaneous spectra of each moment sequentially along the time axis, and using the energy amplitude as the third dimension display parameter, yields the complete time-frequency energy distribution spectrum. In the distribution spectrum, a persistent, slow trend change usually manifests as energy concentration in the low-frequency region, while the pulse signal caused by the instantaneous opening and closing of microcracks often appears as a sudden, temporally concentrated bright spot or energy ridge in the high-frequency region.

[0031] For the time-frequency energy distribution spectrum, a sliding median filter of energy intensity is performed along the frequency axis to extract the time-frequency energy clustering band with energy intensity higher than the median filter benchmark. Based on the chirped modulation characteristics of the microcrack opening and closing pulse in the time-frequency domain, a continuity constraint of the instantaneous frequency change rate is applied to the time-frequency energy clustering band to identify the chirped pulse response atom sequence corresponding to the time-frequency energy clustering band, and the initial instantaneous time-frequency ridge line of each chirped pulse response atom is generated. The initial instantaneous time-frequency ridge is used as the initial value of the local narrowband constraint for chirped mode decomposition. The evolution trajectory of the difference factor is iteratively stripped in the time-frequency domain. In each iteration, the frequency direction of the central ridge of the chirped mode and the local bandwidth parameters are automatically updated according to the current time-frequency energy distribution of the residual signal. The chirped mode components corresponding to different chirped impulse response atoms are stripped one by one until the time-frequency energy distribution of the residual signal no longer has chirped modulation characteristics, thus forming a set of chirped mode components. Intermittency degree discrimination is performed on each chirped mode component in the set of chirped mode components, the energy activation duty cycle of each chirped mode component on the time axis is calculated, and candidate leakage pulse mode components with energy activation duty cycles lower than the intermittency discrimination threshold and energy activation periods showing sudden concentrated characteristics are screened out from the set of chirped mode components. The consistency of the mechanical opening and closing response of the cross-pipe section is verified for the candidate leakage pulse mode components. The candidate leakage pulse mode component with the maximum time domain overlap with the opening and closing time window corresponding to the thermo-mechanical coupling excitation of the welding pulse on the cross-pipe section is determined as the non-stationary leakage pulse characteristic mode component that is finally stripped out.

[0032] It should be noted that after obtaining the time-frequency energy distribution spectrum, in order to locate the chirped pulse response atoms related to microcrack initiation and closure, a sliding median filter is first applied along the frequency direction to adaptively generate a median energy reference surface that varies with frequency. The portion of energy values ​​in each time-frequency unit that exceeds this reference is extracted as the time-frequency energy cluster. Considering that the microcrack initiation and closure behavior is mechanically manifested as a rapid opening-closing process of the crack surface under transient stress, the resulting flow pulse usually exhibits a monotonically changing frequency with the pulse duration in the time-frequency domain, i.e., chirped modulation characteristics. Therefore, during identification, not only are candidate clusters required to have concentrated energy, but also their instantaneous frequency change rate along the time axis must be subject to a continuity constraint to eliminate false clusters with abrupt or random frequency jumps. The finally identified effective clusters correspond to the chirped pulse response atom sequence, and the frequency direction of each atom along the time axis is extracted as the initial instantaneous time-frequency ridge.

[0033] The initial ridges described above provide the local narrowband constraint starting point for chirped mode decomposition. During the iterative stripping phase, these initial ridges serve as the center, limiting the search and extraction of signal components to a narrow frequency band near the ridges, thus avoiding energy aliasing between different modes. After each iteration strips out a chirped mode component, the time-frequency energy distribution of the residual signal is recalculated, and the central ridge frequency direction and search bandwidth of the remaining components to be decomposed are updated accordingly before proceeding to the next stripping iteration. This process is repeated until no energy accumulation structure satisfying the chirped modulation characteristic conditions exists in the residual signal. The resulting stripping results constitute the set of chirped mode components.

[0034] Since the dataset may contain leakage pulse components caused by microcrack initiation and closure, as well as occasional transient components caused by other factors such as solenoid valve operation and water pressure fluctuations, a two-step screening process is required. The first step is to determine the degree of intermittency by calculating the percentage of time each modal component is in an energy-activated state throughout the entire observation period, i.e., the energy activation duty cycle. Microcrack initiation and closure pulses typically have a low duty cycle and a sudden, concentrated activation period because they only appear at the moment the welding pulse is applied. The second step is to verify the consistency of the mechanical initiation and closure response of the cross-connector section. Using the welding current pulse triggering time as a benchmark, the initiation and closure time window of the cross-connector section under thermo-mechanical coupling excitation in each welding cycle is determined. The energy activation period of each candidate modal component is compared with the initiation and closure time window in the time domain. The one with the highest degree of overlap is determined to be the non-stationary leakage pulse characteristic modal component that has a causal relationship with the microcrack initiation and closure of the cross-connector section.

[0035] Specifically, S5 is: The non-stationary leakage pulse characteristic mode component is framed for energy spectral density estimation along the time axis. The time-domain amplitude waveform of the non-stationary leakage pulse characteristic mode component is divided into multiple energy monitoring frames that are continuous and partially overlapping. The normalized energy density within each frame is calculated to construct a pulse energy time distribution sequence that characterizes the non-uniformity of the leakage pulse energy distribution along the welding cycle time axis. Based on the pulse energy time distribution sequence, an information entropy measurement mechanism is introduced to quantify the non-stationary intermittent characteristics of leakage pulse energy release. The normalized energy density of each energy monitoring frame is regarded as a probability distribution measure. The local information entropy of leakage pulse energy distribution is calculated frame by frame along the time axis, and the local information entropy of all energy monitoring frames is statistically aggregated to obtain the intermittent energy entropy that reflects the strength of the intermittency and the degree of temporal disorder of leakage pulse energy release. Retrieve a pre-constructed leakage confidence interval, compare the intermittent energy entropy with the upper bound of the leakage confidence interval using a threshold comparison, and determine whether the intermittent energy entropy exceeds the upper bound of the leakage confidence interval; When the intermittent energy entropy exceeds the upper bound of the leakage confidence interval, the leakage state flag is triggered and the occurrence time of the energy peak of each energy monitoring frame in the non-stationary leakage pulse characteristic mode component is extracted back, and the leakage pulse occurrence time index set is constructed. Based on the temporal correlation between each time in the leakage pulse occurrence time index set and the opening and closing time window of the cross-connection section of the U-shaped cooling water channel under the thermo-mechanical coupling excitation of the welding pulse, after confirming that the leakage pulse occurrence sequence and the cross-connection section opening and closing sequence have causal synchronization characteristics, the leakage location is marked in the cross-connection section area.

[0036] It should be noted that a significant physical characteristic of microcrack leakage is that it is not a continuous leakage, but rather an intermittent opening and closing under the thermo-mechanical coupling of the welding pulse. This results in the leakage pulse signal exhibiting an intermittent distribution pattern of sudden appearance, brief duration, and then disappearance on the time axis. To measure this characteristic, this step introduces an information entropy measurement mechanism. Specifically, the time waveform of the non-stationary leakage pulse characteristic mode component is divided into a series of partially overlapping energy monitoring frames. After calculating the normalized energy density frame by frame, the energy density of each frame within the entire observation period is considered as a probability distribution. The local information entropy reflecting the degree of energy concentration in each frame is calculated frame by frame. Then, the local information entropies of all frames are aggregated to obtain the intermittent energy entropy. The physical meaning of the entropy value is as follows: if the leakage pulse energy is highly concentrated in a few frames on the time axis, the energy distribution of each frame is extremely uneven, and the entropy value is low, indicating that the leakage exhibits significant intermittent burst characteristics; conversely, if the signal is continuous and stable background noise, and the energy is evenly distributed among the frames, the entropy value is high. Therefore, the magnitude of the intermittent energy entropy directly reflects the intermittent intensity and temporal disorder of the leakage pulse energy release. The lower the entropy value, the stronger the intermittency, and the more it conforms to the characteristic mode of microcrack opening and closing leakage.

[0037] The calculated intermittent energy entropy is compared with a pre-constructed leakage confidence interval. This leakage confidence interval is a healthy baseline range established based on statistical analysis of the frequency band energy fluctuation records corresponding to the characteristic mode components of non-stationary leakage pulses during the historical normal operation of the same transformer. It characterizes the natural intermittency level of the energy distribution in this frequency band under leakage-free conditions. When the intermittent energy entropy is lower than the lower bound of the leakage confidence interval (or exceeds the upper bound depending on the specific statistical construction method), it indicates that the intermittency of the current signal has exceeded the reasonable fluctuation range of the leakage-free state, meeting the leakage judgment condition and triggering the leakage status flag to be set.

[0038] Finally, after determining the existence of a leak, the leak location needs to be confirmed and marked: Simultaneously with triggering the leak marker, the times when energy peaks occur in each energy monitoring frame are extracted back to construct an index set of leak pulse occurrence times. Each time in the index set is then compared in the time domain with the opening and closing time window of the cross-connector section under the thermo-mechanical coupling excitation of the welding pulse. Here, the opening and closing time window refers to the effective period of force opening and closing of the crack surface of the cross-connector section, determined by considering the delays in heat conduction and mechanical stress transmission, starting from the trigger time of each welding current pulse. If the occurrence time of the leak pulse consistently falls within this opening and closing time window across multiple welding cycles and exhibits cycle-level repeatability, it indicates a stable causal synchronization relationship between the leak pulse and the welding pulse. This confirms that the leak source is located in the cross-connector section directly affected by the thermo-mechanical coupling excitation of the welding pulse, thus ultimately marking the leak location in this area. This eliminates the possibility of false alarms caused by occasional disturbances in other parts, ensuring the reliability of the location conclusion.

[0039] In addition, the method further includes the following steps: High-order cumulant slice spectrum analysis was performed on the characteristic modal components of the non-stationary leakage pulse. Diagonal slices of the third-order cumulant corresponding to the transient impact of microcrack opening and closing were extracted from the characteristic modal components of the non-stationary leakage pulse to suppress the fundamental frequency harmonic interference of the welding current and obtain the non-Gaussian impact characteristic spectrum of the leakage pulse. Based on the non-Gaussian impact characteristic spectrum of the leakage pulse, the characteristic spectral peaks with local maxima are located along the frequency axis, and the frequency values ​​and their relative kurtosis change rates corresponding to each characteristic spectral peak are extracted to form a sparse characteristic vector in the frequency domain of the leakage impact. The sparse feature vector of leakage impact frequency domain is cross-correlated and matched with the family of pre-calibrated natural mode frequencies of the cross-pipe section under the thermo-mechanical coupling excitation of welding pulse. After eliminating the pseudo leakage spectrum peaks that coincide with the structural resonance mode, the subset of leakage impact feature frequencies that belong only to the transient energy release of crack opening and closing is selected. The relative kurtosis of each frequency component in the leakage impact characteristic frequency subset is arranged according to the welding cycle time sequence to form a crack initiation and closure frequency domain evolution matrix. The crack initiation and closure frequency domain evolution matrix is ​​recursively quantitatively analyzed along the time dimension to determine the recursion rate index that describes the degree of determinism of the recursive appearance of crack initiation and closure frequency components over time. The recursion rate index is used as a quantitative representation of the regularity of microcrack opening and closing behavior in the cross-connection section, and is updated in real time to the leakage state marker to complete the final identification from leakage pulse mode to crack dynamic recursive characteristics.

[0040] It should be noted that in actual welding environments, the transformer body is subjected to the coupling effect of strong electromagnetic fields and complex mechanical vibrations. Although the non-stationary leakage pulse characteristic modal components extracted have been determined to be related to leakage, their frequency domain composition may still contain spurious impact components caused by crosstalk of fundamental harmonics of welding current or structural resonance response. If the unfiltered modal components are directly used as leakage characteristics, false warnings may occur in long-term monitoring due to harmonic fluctuations in the operating conditions. Therefore, the purpose of this embodiment is to further extract the frequency domain impact characteristics directly related to the transient energy release of microcrack opening and closing from the modal components, and to quantify their occurrence patterns in the time series at the dynamic level, so as to improve the confidence of leakage identification. Specifically, the high-order cumulant slice spectrum analysis approach leverages the inherent insensitivity of the third-order cumulant to Gaussian noise and symmetrical interference. It processes the characteristic modal components of the non-stationary leakage pulse, extracting diagonal slices of its third-order cumulant. This operation effectively suppresses interference from symmetrical periodic signals such as the fundamental frequency and harmonics of the welding current while preserving the asymmetric transient impact components caused by microcrack initiation and closure, thus obtaining a relatively pure non-Gaussian impact characteristic spectrum of the leakage pulse. In the characteristic spectrum, characteristic spectral peaks with local maxima are located along the frequency axis, each peak corresponding to a concentrated frequency band of leakage impact energy. The frequency values ​​of each characteristic spectral peak are extracted, and the local rate of change of its kurtosis relative to adjacent frequency bands is calculated, thereby constructing a sparse characteristic vector of the leakage impact in the frequency domain. Since the physical process of microcrack initiation and closure has instantaneous energy release characteristics, its manifestation in the frequency domain is usually a finite number of sparsely distributed narrowband impact peaks, while background vibration interference often has a relatively flat and diffuse spectral distribution. Sparse constraints can initially filter out non-impact background components.

[0041] Then, the sparse eigenvectors of the leakage impact frequency domain are cross-correlated and matched with the pre-calibrated family of inherent modal frequencies of the cross-pipe segment under the thermo-mechanical coupling excitation of the welding pulse. The inherent modal frequency family refers to a set of inherent structural resonance frequencies possessed by the cross-pipe segment as a mechanical component with specific geometric dimensions and materials under thermal stress and welding pulse excitation, which can be obtained in advance through offline impact modal testing or finite element simulation. Matching the leakage impact spectral peak frequencies with the inherent modal frequency family aims to eliminate pseudo-leakage spectral peaks that, although exhibiting impact characteristics, actually originate from structural resonance responses, retaining only spectral peaks whose frequencies do not match the structural modes as a subset of leakage impact characteristic frequencies. The frequency components in this subset are directly attributed to the stress wave energy released by the instantaneous opening and closing of the crack surface. The relative kurtosis change rate corresponding to each frequency component in the subset is arranged according to the welding cycle time sequence to construct a crack opening and closing frequency domain evolution matrix. Recursive quantitative analysis is performed on the matrix along the time dimension, the principle of which is to detect the deterministic pattern of recursive recurrence of the leakage impact frequency components in the evolution matrix between different welding cycles. Since the initiation and closure of microcracks are driven by the periodic thermo-mechanical coupling of welding pulses, the actual leakage impact should exhibit a recursive occurrence pattern synchronized with the welding cycle, while random interference does not possess this characteristic. The recursion rate index obtained from the analysis describes the degree of determinism of the recursive occurrence of crack initiation and closure frequency components over time: the higher the recursion rate, the more obvious the regular microcrack initiation and closure behavior in the signal.

[0042] Finally, the recursion rate index is used as a quantitative representation of the regularity of microcrack opening and closing behavior in the cross-connection section, and is updated in real time to the leakage status marker, thereby completing the final identification from leakage pulse mode components to crack dynamic recursive characteristics. This identification result can provide a more sufficient basis for leakage degree assessment and maintenance decision-making.

[0043] In addition, the method further includes the following steps: The multidimensional flow observation matrix is ​​extended along the welding cycle dimension as the time axis. The three-dimensional array formed by stacking the window flow imbalance vectors corresponding to multiple consecutive welding cycles is directly designated as a flow-imbalance-cycle third-order tensor, where the three dimensions correspond to the branch node flow direction deviation, the flow imbalance amplitude and the welding cycle number, respectively. The welding current effective value sequence and water inlet temperature sequence obtained in step S3 are synchronously acquired, and the instantaneous amplitude sequence of the non-stationary leakage pulse characteristic mode component stripped in step S4 are spliced ​​together according to the welding cycle number to form a working condition auxiliary matrix, which serves as the supervisory and guiding input for coupled tensor decomposition. The sparse-constrained coupled tensor decomposition is performed on the flow-imbalance-cycle third-order tensor. The working condition auxiliary matrix is ​​used as a coupling factor to participate in the decomposition iteration. By applying the kernel tensor sparse regularization term, only the coupled modes with physical meaning are retained in the constraint decomposition result. The output is an approximate tensor consisting of the sum of the outer products of multiple rank-1 components and the coupled kernel tensor conjugate thereto. From each modal slice of the coupled kernel tensor, the proportion of non-zero elements in the slice is calculated sequentially, and the non-stationary leakage pulse characteristic modal component obtained in step S4 is called to perform waveform consistency cross-correlation screening with each modal slice along the welding cycle dimension to separate the leakage characteristic kernel tensor that has the highest correlation with the time domain waveform of the non-stationary leakage pulse mode. The column dimension sparsity of the leakage feature nucleus tensor is calculated. After expanding the leakage feature nucleus tensor into a matrix along the welding cycle dimension, the number of elements in each column whose absolute amplitude value is lower than the preset sparsity threshold is counted. After normalization, it is used as the leakage nonlinear interaction intensity index of multi-physics coupling under the welding cycle.

[0044] It should be noted that the unbalanced quantity time-series stack matrix, with the welding cycle number as the row index and the branch node flow direction deviation as the column vector, has already been obtained. Essentially, this matrix is ​​a two-dimensional data structure formed by vertically stacking the flow unbalance vectors corresponding to each welding cycle window. Based on this, using the welding cycle number as the third-dimensional index, the two-dimensional matrix can be directly extended into a three-dimensional data structure. The first dimension corresponds to the branch node flow direction deviation, the second dimension to the flow unbalance magnitude, and the third dimension to the welding cycle number. This is a flow-unbalance-cycle third-order tensor, which fully preserves the structured information of the flow unbalance state at different branch nodes within each welding cycle.

[0045] Furthermore, to guide the tensor decomposition process towards the physical direction related to leakage, an auxiliary operating condition matrix needs to be constructed simultaneously. Specifically, the welding current RMS sequence and inlet water temperature sequence acquired synchronously in step S3, along with the instantaneous amplitude sequence of the non-stationary leakage pulse characteristic mode components extracted in step S4, are aligned according to the welding cycle number and then concatenated to form a coupling factor matrix. Each row of the matrix corresponds to a welding cycle, and each column represents the current intensity, cooling medium temperature, and pulse response amplitude of the confirmed leakage source under that cycle.

[0046] By performing sparse-constrained coupled tensor decomposition on the aforementioned third-order tensor, the working condition auxiliary matrix is ​​used as a coupling factor in the iterative optimization. During the decomposition process, a sparse regularization term is applied to the solved kernel tensor, forcing the decomposition result to retain only a few non-zero components. This constraint has a clear physical meaning: in reality, the multi-field coupling effect related to leakage is only significant at specific welding cycles and specific branch nodes. This is reflected in the tensor decomposition result as the corresponding kernel tensor should exhibit a sparse distribution. Background fluctuation components unrelated to leakage are naturally filtered out during iteration because they do not possess this sparse structure. After decomposition, the output is an approximate tensor consisting of the sum of the outer products of multiple rank-one components, and its conjugate coupled kernel tensor. Each slice of the coupled kernel tensor represents the distribution of different coupling modes in each dimension.

[0047] Next, the leakage characteristic nucleus tensor is extracted from the coupling kernel tensor: Each modal slice of the coupling kernel tensor is sequentially traversed, and the proportion of non-zero elements in each slice is statistically analyzed. Simultaneously, the non-stationary leakage pulse characteristic modal components confirmed in step S4 are called, and their time-domain waveforms are cross-correlated with the time series obtained by expanding each modal slice along the welding cycle dimension. The slice with the highest waveform consistency is selected. The physical meaning of a slice is that the change pattern of the coupling mode it represents in the time dimension highly matches the confirmed leakage pulse response. Therefore, it can be determined that the nucleus tensor corresponding to this slice is the leakage characteristic nucleus tensor that only carries leakage-related multi-field coupling information.

[0048] Finally, column-dimensional sparsity calculations are performed on the obtained leakage feature nucleus tensor to generate a leakage nonlinear interaction intensity index. The leakage feature nucleus tensor is expanded into a matrix along the welding cycle dimension. The number of elements with absolute amplitude values ​​below a preset sparsity threshold is counted column by column and normalized. The resulting value characterizes the nonlinear interaction intensity of the multiphysics coupling to the leakage response at that welding cycle. A smaller index value indicates sparser coupling and more concentrated leakage features within that cycle; conversely, a larger value indicates dispersed multiphysics coupling and relatively weakened leakage features. The intensity index can be updated in real-time to the leakage status marker, providing a quantitative reference for assessing the leakage degree and predicting its development trend.

[0049] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A cooling water circuit structure for a medium-frequency resistance welding transformer, comprising a transformer body, characterized in that: The transformer body includes an external protective shell, a water circuit board embedded inside the protective shell, and a transformer heating core component. The water channel plate is a rigid metal plate structure. The left and right sides of the water channel plate are integrally formed with symmetrical wing mounting structures that are mirror-distributed. The symmetrical wing mounting structures are wing-shaped mounting bosses that extend horizontally outward from the side of the water channel plate. The wing-shaped mounting bosses have mounting holes that are through-holes along the thickness direction for locking and fixing with the assembly base. The interior of the water circuit board is formed by deep hole processing to create a U-shaped cooling water circuit for the flow of cooling water. The U-shaped cooling water circuit consists of a longitudinal main water circuit extending along the vertical central axis of the water circuit board and two sets of transverse branch water circuits arranged vertically and parallel to each other. The longitudinal main water circuit and the two sets of transverse branch water circuits intersect to form a closed-loop U-shaped circulating cooling channel. The lower lateral branch of the U-shaped cooling water circuit is provided with an arc-shaped cross-connection section. The cross-connection section crosses laterally across the mounting hole area on the symmetrical wing mounting structure, so that the cooling water circuit and the transformer mounting and fixing structure are integrated into one unit.

2. The cooling water circuit structure of a medium-frequency resistance welding transformer according to claim 1, characterized in that, The connection point between the longitudinal main waterway and the two sets of transverse branch waterways adopts a smooth arc transition structure.

3. The cooling water circuit structure of a medium-frequency resistance welding transformer according to claim 1, characterized in that, The two wing-shaped mounting bosses of the symmetrical wing mounting structure are strictly symmetrically distributed along the vertical central axis of the water channel plate, and the thickness of the wing-shaped mounting bosses is consistent with the thickness of the main body of the water channel plate.

4. The cooling water circuit structure of a medium-frequency resistance welding transformer according to claim 1, characterized in that, The inner wall of the cross-connector section is a smooth curved surface, and a safety clearance is reserved between the cross-connector section and the mounting hole to avoid structural interference between the water channel and the mounting hole, while ensuring smooth flow of cooling water within the cross-connector section.

5. A leakage detection method for the cooling water circuit structure of a medium-frequency resistance welding transformer, applied to the cooling water circuit structure of the medium-frequency resistance welding transformer as described in any one of claims 1 to 4, characterized in that, Includes the following steps: S1. When the medium frequency resistance welding transformer is under load welding condition and the cooling water circuit is circulating, the flow time sequence signal of the inlet end and the outlet end of the upper and lower transverse branch is collected simultaneously to construct a multi-dimensional flow observation matrix that reflects the topological constraint relationship of the water circuit. S2. Based on the multidimensional flow observation matrix, calculate the normalized leakage flow difference factor to characterize the leakage degree of the lower transverse branch; S3. Compare the trend deviation of the normalized leakage flow difference factor with the adaptive leakage threshold cloud map based on transformer operating condition identification to generate the evolution trajectory of the difference factor to describe the dynamic diffusion behavior of leakage. S4. Perform modal decomposition on the evolution trajectory of the difference factor to extract the non-stationary leakage pulse characteristic modal components induced by the opening and closing of microcracks in the cross-pipe section. S5. When the intermittent energy entropy of the non-stationary leakage pulse characteristic mode component exceeds the preset leakage confidence interval, it is determined that a leakage has occurred in the cooling water circuit and the leakage location is marked in the cross-connection section area.

6. The leakage detection method according to claim 5, characterized in that, Specifically, S1 is: At the starting moment of synchronously triggering the welding pulse of the medium-frequency resistance welding transformer, equal-duration segmented windowing is performed on the original flow time-series signals output by the flow sensor at the water inlet end, the flow sensor at the outlet end of the upper horizontal shunt branch, and the flow sensor at the outlet end of the lower horizontal shunt branch, to obtain the water inlet end flow window sequence, the upper branch flow window sequence, and the lower branch flow window sequence that are synchronously locked with a single welding cycle; The flow mean value feature and the flow fluctuation kurtosis feature within the window are respectively extracted from the water inlet end flow window sequence, the upper branch flow window sequence, and the lower branch flow window sequence, and the flow mean value feature and the flow fluctuation kurtosis feature of each window are feature-stitched according to the time-series index to form the water inlet end characteristic flow, the upper branch characteristic flow, and the lower branch characteristic flow that characterize the flow states of each measuring point; According to the flow resistance topological constraint relationship of the Chinese character-shaped cooling water path at the cross-connecting pipe section node, the water inlet end characteristic flow, the upper branch characteristic flow, and the lower branch characteristic flow are substituted into the matrix expression framework based on the fluid mechanics continuity theorem point by point along the time axis, and the window flow imbalance vector reflecting the degree of flow distribution imbalance of each branch within the window is calculated window by window; The window flow imbalance vectors of continuous multiple welding cycles are longitudinally stacked in chronological order to construct an imbalance time-series stack matrix with the welding cycle number as the row index and the branch node flow deviation as the column vector; The imbalance time-series stack matrix is subjected to element-by-element differential operation with the flow resistance topological constraint matrix under the normal state of the pre-calibrated water path to obtain a multi-dimensional flow observation matrix reflecting the degree of deviation of the flow distribution of the Chinese character-shaped water path from the topological constraint under the loaded condition.

7. The leakage detection method according to claim 5, characterized in that, The specific content of S2 is as follows: From the multi-dimensional flow observation matrix, the flow distribution deviation vectors corresponding to each window are extracted row by row according to the welding cycle index to form a deviation vector sequence reflecting the flow deviation amplitude of each branch of the Chinese character-shaped water path within a single welding cycle; According to the flow resistance topological connection relationship of the Chinese character-shaped cooling water path at the cross-connecting pipe section node, a Kirchhoff node association constraint mode with the water inlet end confluence node, the upper branch shunt node, and the lower branch cross-connecting pipe section shunt node as topological units is constructed, and each deviation vector in the deviation vector sequence is sequentially substituted into the Kirchhoff node association constraint mode to calculate and obtain a node flow constraint residual vector characterizing the node flow conservation residual; The node flow constraint residual vectors corresponding to continuous multiple welding cycles are arranged in chronological order to generate a node flow constraint residual matrix, and according to the branch correspondence relationship between the lower horizontal shunt branch and the cross-connecting pipe section in the flow resistance topology, local flow resistance sensitivity screening is applied to the node flow constraint residual matrix to separate the lower branch leakage residual contribution component matrix that is only related to the leakage source strength of the lower horizontal shunt branch; The singular feature direction extraction of the leakage mode is performed on the lower branch leakage residual contribution component matrix to obtain a main leakage feature direction vector that can maximize the characterization of the leakage amplitude fluctuation of the lower horizontal shunt branch; The main leakage characteristic direction vector is projected onto the direction of the average total flow rate at the inlet, and normalized and scaled using the continuous constraint of the overall water flow, and finally the normalized leakage flow rate difference factor is calculated.

8. The leakage detection method according to claim 5, characterized in that, Specifically, S3 is: The welding current waveform signal and the inlet water temperature signal of the T-shaped cooling water circuit are synchronously acquired during the welding process of the transformer under load. The effective value of the welding current and the inlet water temperature value are used to form a binary group of the current operating conditions. Based on the pre-established adaptive leakage threshold cloud map, the upper limit and lower limit of the adaptive leakage threshold of the operating conditions that match the binary group of the current operating conditions are found and extracted to form the adaptive leakage threshold range under the current operating conditions. The time-series values ​​of the normalized leakage flow difference factor on multiple consecutive welding cycles are compared with the membership degree of the adaptive leakage threshold interval at the corresponding cycle time. It is determined whether the normalized leakage flow difference factor of each cycle falls within or outside the adaptive leakage threshold interval. The deviation direction and deviation magnitude value of the difference factor of each cycle relative to the center of the adaptive leakage threshold interval are extracted to form a leakage deviation state sequence. The deviation amplitude values ​​of each cycle in the leakage deviation state sequence are subjected to sliding weighted cumulative deviation trend enhancement processing along the time axis. The increment of the same-direction deviation amplitude between adjacent cycles is accumulated as the trend cumulative deviation, while suppressing the instantaneous reverse deviation caused by measurement noise, and generating a trend cumulative deviation sequence that reflects the continuous change trend of the leakage deviation degree. The trend cumulative deviation sequence is mapped to a two-dimensional deviation phase plane with the welding cycle number as the horizontal axis and the trend cumulative deviation as the vertical axis. The points on the two-dimensional deviation phase plane are connected in time evolution order to form an initial deviation phase trajectory describing the dynamic diffusion behavior of the normalized leakage flow difference factor relative to the adaptive leakage threshold range. The initial deviation phase trajectory is smoothed and the trend direction feature is extracted based on phase space reconstruction. The high-frequency reciprocating fluctuation component caused by instantaneous disturbance of the operating condition is removed from the initial deviation phase trajectory, and the dominant trend skeleton that can reflect the consistency of leakage diffusion direction and the change law of diffusion rate is retained to generate the difference factor evolution trajectory.

9. The leakage detection method according to claim 5, characterized in that, Specifically, S4 is: The time-frequency energy distribution spectrum of the evolution trajectory of the difference factor is constructed along the welding cycle time axis, transforming the evolution trajectory of the difference factor from a single amplitude-time domain to a three-dimensional joint characterization space of time-frequency-energy, thereby obtaining the time-frequency energy distribution spectrum. For the time-frequency energy distribution spectrum, a sliding median filter of energy intensity is performed along the frequency axis to extract the time-frequency energy clustering band with energy intensity higher than the median filter benchmark. Based on the chirped modulation characteristics of the microcrack opening and closing pulse in the time-frequency domain, a continuity constraint of the instantaneous frequency change rate is applied to the time-frequency energy clustering band to identify the chirped pulse response atom sequence corresponding to the time-frequency energy clustering band, and the initial instantaneous time-frequency ridge line of each chirped pulse response atom is generated. The initial instantaneous time-frequency ridge is used as the initial value of the local narrowband constraint for chirped mode decomposition. The evolution trajectory of the difference factor is iteratively stripped in the time-frequency domain. In each iteration, the frequency direction of the central ridge of the chirped mode and the local bandwidth parameters are automatically updated according to the current time-frequency energy distribution of the residual signal. The chirped mode components corresponding to different chirped impulse response atoms are stripped one by one until the time-frequency energy distribution of the residual signal no longer has chirped modulation characteristics, thus forming a set of chirped mode components. Intermittency degree discrimination is performed on each chirped mode component in the set of chirped mode components, the energy activation duty cycle of each chirped mode component on the time axis is calculated, and candidate leakage pulse mode components with energy activation duty cycles lower than the intermittency discrimination threshold and energy activation periods showing sudden concentrated characteristics are screened out from the set of chirped mode components. The consistency of the mechanical opening and closing response of the cross-pipe section is verified for the candidate leakage pulse mode components. The candidate leakage pulse mode component with the maximum time domain overlap with the opening and closing time window corresponding to the thermo-mechanical coupling excitation of the welding pulse on the cross-pipe section is determined as the non-stationary leakage pulse characteristic mode component that is finally stripped out.

10. The leakage detection method according to claim 5, characterized in that, Specifically, S5 is: The non-stationary leakage pulse characteristic mode component is framed for energy spectral density estimation along the time axis. The time-domain amplitude waveform of the non-stationary leakage pulse characteristic mode component is divided into multiple energy monitoring frames that are continuous and partially overlapping. The normalized energy density within each frame is calculated to construct a pulse energy time distribution sequence that characterizes the non-uniformity of the leakage pulse energy distribution along the welding cycle time axis. Based on the pulse energy time distribution sequence, an information entropy measurement mechanism is introduced to quantify the non-stationary intermittent characteristics of leakage pulse energy release. The normalized energy density of each energy monitoring frame is regarded as a probability distribution measure. The local information entropy of leakage pulse energy distribution is calculated frame by frame along the time axis, and the local information entropy of all energy monitoring frames is statistically aggregated to obtain the intermittent energy entropy that reflects the strength of the intermittency and the degree of temporal disorder of leakage pulse energy release. Retrieve a pre-constructed leakage confidence interval, compare the intermittent energy entropy with the upper bound of the leakage confidence interval using a threshold comparison, and determine whether the intermittent energy entropy exceeds the upper bound of the leakage confidence interval; When the intermittent energy entropy exceeds the upper bound of the leakage confidence interval, the leakage state flag is triggered and the occurrence time of the energy peak of each energy monitoring frame in the non-stationary leakage pulse characteristic mode component is extracted back, and the leakage pulse occurrence time index set is constructed. Based on the temporal correlation between each time in the leakage pulse occurrence time index set and the opening and closing time window of the cross-connection section of the U-shaped cooling water channel under the thermo-mechanical coupling excitation of the welding pulse, after confirming that the leakage pulse occurrence sequence and the cross-connection section opening and closing sequence have causal synchronization characteristics, the leakage location is marked in the cross-connection section area.

Images