Intelligent assembly and welding manufacturing method and system for open press bed

Abstract

The present invention provides an open press machine bed intelligent assembly and welding manufacturing method and system, which relates to the open press machine bed assembly and welding technical field, including dividing the machine bed into a plurality of independent functional modules according to the structural characteristics of the open press machine bed, positioning and clamping each functional module to obtain an assembled open press machine bed; establishing a welding sequence optimization model for the key force-bearing structure of the open press machine bed, and obtaining the optimal welding sequence of the weld by solving the welding sequence optimization model; welding the assembled open press machine bed using the optimal welding sequence, and in the welding process, using phased array ultrasonic imaging technology to identify the weld type and defect characteristics, outputting ultrasonic imaging results of welding defects, predicting the expansion evolution behavior of defects according to the intelligent detection results, forming a quantitative defect evaluation system, and realizing multi-information fusion intelligent evaluation of welding quality.

Description

Technical Field

[0001] The invention relates to an open press machine bed assembly and welding technology, and in particular to an open press machine bed intelligent assembly and welding manufacturing method and system. Background Art

[0002] The bed of an open press is a key component of a press, and its manufacturing quality directly affects the overall performance of the press. The traditional manufacturing process of an open press bed mainly uses manual assembly and welding, which has problems such as low assembly accuracy, large welding deformation, and unstable welding quality, making it difficult to meet the accuracy and quality requirements of high-end presses. In recent years, with the development of technologies such as industrial Internet and artificial intelligence, intelligent manufacturing technology has been widely used, providing new ideas and methods for the manufacturing of open press beds.

[0003] At present, the intelligent manufacturing method for large and complex welded structural parts such as open press beds is still immature. There is a lack of modular assembly methods specifically for the structural characteristics of open press beds, which makes it difficult to achieve accurate and fast positioning and clamping; there is a lack of welding sequence optimization methods that consider the stress characteristics of key force-bearing structures, which makes it difficult to effectively control welding deformation; there is a lack of intelligent welding quality evaluation methods that integrate multi-source sensor information, which makes it difficult to achieve quantitative analysis and prediction of defects. Therefore, it is urgent to develop a new intelligent assembly and welding manufacturing method for open press beds to improve the manufacturing efficiency and quality of open press beds. Summary of the invention

[0004] The embodiment of the present invention provides an open press machine bed intelligent assembly and welding manufacturing method and system, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] Provided is an intelligent assembly and welding manufacturing method for an open press bed, comprising:

[0007] According to the structural characteristics of the open press bed, the machine bed is divided into several independent functional modules, and an independent assembly process is formulated for each functional module, including sub-module division, positioning and clamping, welding stacking and precision measurement. The machine bed is placed on a pre-set intelligent fixture, and the displacement sensing unit of the intelligent fixture is used to scan the actual spatial position data of the machine bed. The scanned actual spatial position data is compared with the design model of the machine bed, and the adjustment amount of the support point is calculated. According to the adjustment amount, each functional module is positioned and clamped to obtain an assembled open press bed;

[0008] Aiming at the key force-bearing structure of the open press bed, a welding sequence optimization model of the weld is established, the maximum displacement deformation in the characteristic direction of the key force-bearing structure is taken as the optimization target, the geometric constraints and process constraints of the weld are taken as the constraint conditions, and the optimal welding sequence of the weld is obtained by solving the welding sequence optimization model;

[0009] The assembled open press bed is welded using the optimal welding sequence. During the welding process, phased array ultrasonic imaging technology is used to identify the weld type and defect characteristics, and the ultrasonic imaging results of the welding defects are output. The machine vision algorithm is used to perform intelligent detection on the ultrasonic imaging results of the welding defects. The expansion and evolution behavior of the defects is predicted based on the intelligent detection results, forming a quantitative defect evaluation system and realizing multi-information fusion intelligent evaluation of welding quality.

[0010] In an optional embodiment,

[0011] The displacement sensing unit of the intelligent fixture scans the actual spatial position data of the machine tool bed, compares the scanned actual spatial position data with the design model of the machine tool bed, calculates the adjustment amount of the support point, and positions and clamps each functional module according to the adjustment amount, so that the assembled open press bed includes:

[0012] Use a large-size three-dimensional coordinate measuring machine to measure the geometric accuracy of the machine tool bed and obtain the point cloud data of the machine tool bed that reflects the actual processing status of the bed;

[0013] Import the point cloud data of the machine tool bed into the computer, generate the actual processing model of the machine tool bed through the three-dimensional reconstruction technology, compare the actual processing model of the machine tool bed with the design model of the machine tool bed, extract and quantify the geometric deviation between the actual processing model of the machine tool bed and the design model of the machine tool bed, and establish the geometric deviation vector field between the actual processing model of the machine tool bed and the design model of the machine tool bed;

[0014] Based on the geometric deviation vector field of the actual processing model of the machine tool bed and the design model of the machine tool bed, a nonlinear mapping model of the assembly error between the machine tool bed assembly error and the machine tool bed adjustment parameters is constructed, where the machine tool bed adjustment parameters include the posture parameters of each functional module of the machine tool bed and the support parameters of the whole machine bed, and a neural network algorithm is used to train the nonlinear mapping model of the assembly error.

[0015] Obtain the target accuracy index and input the trained assembly error nonlinear mapping model, solve the optimal value of the machine tool bed adjustment parameter through the intelligent optimization algorithm, and generate the optimal value solution of the machine tool bed adjustment parameter;

[0016] By using the high-precision servo drive unit integrated in the intelligent fixture, the optimal value scheme of the machine tool bed adjustment parameters is converted into actual adjustment actions, and the functional modules of the machine tool bed are positioned and clamped. It is judged again whether the geometric accuracy indicators of the machine tool bed after adjustment reach the target accuracy. If not, the steps of obtaining the optimal value scheme of the bed adjustment parameters and converting it into actual adjustment actions are repeated iteratively until the geometric accuracy indicators of the machine tool bed after optimization and adjustment reach the target accuracy requirements, and the assembled open press bed is obtained.

[0017] In an optional embodiment,

[0018] Obtain the target accuracy index and input the trained assembly error nonlinear mapping model. Use the intelligent optimization algorithm to solve the optimal value of the machine tool bed adjustment parameters. The optimal value solution for the machine tool bed adjustment parameters includes:

[0019] The core concepts, attributes and semantic relationships in the machine tool field are defined through ontology modeling methods. Natural language processing technology is used to extract machine tool accuracy-related entities, attributes and relationship facts from structured and unstructured data sources. Knowledge fusion technology is used to perform representation learning, disambiguation links and consistency reasoning on multi-source heterogeneous machine tool accuracy knowledge, and a machine tool accuracy knowledge graph is constructed.

[0020] The conceptual nodes and semantic relationships related to causal analysis are extracted from the constructed machine tool accuracy knowledge graph to form causal prior knowledge, introduce domain scope constraints, and construct a machine tool error causal graph through knowledge-guided causal structure learning strategy;

[0021] The error data in the machine tool bed assembly process is obtained to construct training samples. The machine tool error cause-effect diagram is used as prior knowledge. The nonlinear mapping model of the machine tool assembly error is trained through a deep learning algorithm. The nonlinear mapping relationship between the machine tool influencing factors and the assembly error is established. The target accuracy index is input into the trained nonlinear mapping model of the machine tool assembly error. The optimal value of the machine tool bed adjustment parameters is solved through an intelligent optimization algorithm, and the optimal value solution of the machine tool bed adjustment parameters is generated.

[0022] After generating the optimal value solution for machine tool bed adjustment parameters, it also includes:

[0023] The machine tool bed is adjusted according to the optimal value plan of the machine tool bed adjustment parameters, the adjusted machine tool error is tested, the adjusted machine tool error test data is collected, the feature extraction and fusion of multi-source heterogeneous error indicators in the machine tool error test data are performed to obtain the machine tool error pattern vector, and the obtained machine tool error pattern vector is input into the machine tool error causal graph model. By performing a reverse search in the machine tool error causal graph, a causal path from the starting factor to the target error is generated, and the comprehensive influence intensity of the causal path is calculated. The key factors that have the greatest impact on the target error are identified through the factor sorting strategy, the error propagation chain schematic diagram of the key factors is extracted, and an error diagnosis report is automatically generated.

[0024] In an optional embodiment,

[0025] Aiming at the key force-bearing structure of the open press bed, a welding sequence optimization model of the weld is established. The maximum displacement deformation in the characteristic direction of the key force-bearing structure is taken as the optimization target, and the geometric constraints and process constraints of the weld are taken as the constraint conditions. By solving the welding sequence optimization model, the optimal welding sequence of the weld is obtained, including:

[0026] For the key force-bearing structure of the open press bed, the 3D CAD model of the key force-bearing structure is imported, the geometric topological relationship of the weld on the key force-bearing structure is identified, and the finite element analysis software is used to perform thermal-mechanical coupling simulation of the welding process under different welding sequences, predict the post-welding deformation, and post-process the simulation and prediction results. The maximum displacement deformation in the characteristic direction of the key force-bearing structure is extracted as the objective function of the welding sequence optimization model. The geometric constraints and process constraints of the weld are used as constraints, and the initial welding sequence of the weld is obtained by solving the welding sequence optimization model;

[0027] In the process of welding the key stress-bearing structure according to the initial welding sequence, the welding state information is collected in real time through multi-sensor fusion, and the welding state information includes welding parameters, weld geometry, temperature field and strain field;

[0028] Preprocess the collected welding state information, reconstruct the preprocessed welding state information into a state space vector, extract key features reflecting the degree and type of deviation from the state space vector through signal processing methods, construct a deviation discriminant function, calculate the distance between the current welding state and the expected welding state according to the deviation discriminant function, and quantitatively compare it with the set deviation threshold, and judge whether a deviation occurs according to the comparison result;

[0029] If a deviation is detected, the real-time re-optimization operation of the welding sequence is initiated, and the current welding state is evaluated according to the deviation. According to the evaluation result of the current welding state, the constraints in the welding sequence optimization model are dynamically adjusted, and the heuristic search algorithm is used to quickly search for the optimal welding sequence under the new constraints. The difference between the adjusted welding sequence and the original welding sequence is quantitatively evaluated, the adjustment range of the welding sequence and the optimization effect are weighed, and finally the optimal welding sequence of the weld is output.

[0030] In an optional embodiment,

[0031] Extracting the maximum displacement deformation in the key stress-bearing structural characteristic direction as the objective function of the welding sequence optimization model includes:

[0032] The deformation sensitivity coefficient and importance weight are introduced, and the objective function is constructed according to the maximum displacement deformation, root mean square displacement deformation and maximum displacement deformation difference. The formula is as follows:

[0033] ;

[0034] Where x represents the welding sequence vector, n represents the total number of welds, and w i represents the importance weight of the ith weld, s i represents the deformation sensitivity coefficient of the ith weld, λ1 represents the balance coefficient of the maximum displacement deformation, Ω i represents the node set in the area affected by the ith weld, u j (x) represents the displacement deformation of the jth node under the weld sequence x, λ2 represents the balance coefficient of the root mean square displacement deformation, λ3 represents the balance coefficient of the maximum displacement deformation difference, u k (x) represents the displacement deformation of the kth node under the weld sequence x;

[0035] The formula for calculating the distance between the current welding state and the expected welding state based on the deviation discriminant function is as follows:

[0036] ;

[0037] Where D(·) represents the deviation discriminant function, y represents the current welding state feature vector, represents the expected welding state feature vector, α represents the weight coefficient of distance, S represents the covariance matrix of feature vector, and β represents the weight coefficient of similarity.

[0038] In an optional embodiment,

[0039] The assembled open press bed is welded using the optimal welding sequence. During the welding process, phased array ultrasonic imaging technology is used to identify the weld type and defect characteristics, output the ultrasonic imaging results of welding defects, and use machine vision algorithms to perform intelligent detection of the ultrasonic imaging results of welding defects. The expansion and evolution behavior of defects is predicted based on the intelligent detection results, forming a quantitative defect evaluation system to achieve multi-information fusion intelligent evaluation of welding quality, including:

[0040] A two-dimensionally arranged phased array ultrasonic transducer is used to optimize the transmitting end sound field and the receiving end sound field through synthetic aperture focusing, spatial filtering and sparse array methods. Phase calibration and amplitude normalization measures are used to adjust the phase consistency problem and dynamic range problem in the sound field optimization process to obtain the target optimized sound field, and the phased array ultrasonic imaging image is obtained according to the target optimized sound field.

[0041] Based on the obtained phased array ultrasonic imaging images, cluster analysis algorithm and pattern recognition algorithm are used to extract imaging parameter optimization rules from historical phased array ultrasonic imaging images, and imaging parameters are adaptively optimized in combination with expert experience. The imaging parameters include imaging angle range, focusing criteria and imaging image size. After optimization, the best ultrasonic imaging image of weld defects is obtained;

[0042] Extract the amplitude parameters, phase parameters and spectrum parameters of the defect echo signal in the best ultrasonic imaging image, combine the acoustic field model and the defect scattering model, and invert the defect quantitative information corresponding to the best ultrasonic imaging image, including the type, size and orientation information of the defect; at the same time, for three-dimensional defects, reconstruct the three-dimensional defect morphology of the three-dimensional defect through a tomographic imaging method and a cross-section synthesis method;

[0043] The quantitative information of defects and the best ultrasonic imaging image corresponding to the reconstructed three-dimensional defect morphology are converted into a grayscale image, and the morphological and texture features of the defects in the grayscale image are extracted using a machine vision algorithm, and input into a pre-trained convolutional neural network for defect segmentation, recognition and positioning to obtain intelligent detection results of weld defects;

[0044] The defect information in the intelligent detection results is fused with the pre-acquired X-ray detection information, eddy current detection information and magnetic memory detection information. Combined with the material properties of the weld where the defect is located, the welding process parameters and the weld load, an association model between defects and weld structural integrity is established. The quantitative defect information in the intelligent detection results is input into the association model to predict the expansion and evolution behavior of the defects, form a quantitative evaluation system for defects, and finally realize multi-information fusion intelligent evaluation of welding quality.

[0045] According to a second aspect of the embodiments of the present invention,

[0046] Provided is an open press bed intelligent assembly and welding manufacturing system, including:

[0047] The first unit is used to divide the machine bed into several independent functional modules according to the structural characteristics of the open press bed, and formulate an independent assembly process for each functional module, including sub-module division, positioning and clamping, welding stack and precision measurement, and place the machine bed on a pre-set intelligent fixture, scan the actual spatial position data of the machine bed through the displacement sensing unit of the intelligent fixture, compare the scanned actual spatial position data with the design model of the machine bed, calculate the adjustment amount of the support point, and position and clamp each functional module according to the adjustment amount to obtain an assembled open press bed;

[0048] The second unit is used to establish a welding sequence optimization model for the key force-bearing structure of the open press bed, taking the maximum displacement deformation in the characteristic direction of the key force-bearing structure as the optimization target, taking the geometric constraints and process constraints of the weld as the constraint conditions, and obtaining the optimal welding sequence of the weld by solving the welding sequence optimization model;

[0049] The third unit is used to weld the assembled open press bed using the optimal welding sequence. During the welding process, phased array ultrasonic imaging technology is used to identify the weld type and defect characteristics, output the ultrasonic imaging results of welding defects, and use machine vision algorithms to perform intelligent detection on the ultrasonic imaging results of welding defects. The expansion and evolution behavior of defects is predicted based on the intelligent detection results, forming a quantitative defect evaluation system and realizing multi-information fusion intelligent evaluation of welding quality.

[0050] According to a third aspect of the embodiments of the present invention,

[0051] An electronic device is provided, comprising:

[0052] processor;

[0053] a memory for storing processor-executable instructions;

[0054] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0055] A fourth aspect of the embodiments of the present invention is:

[0056] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0057] In this embodiment, fast and accurate positioning and clamping are achieved through intelligent fixtures, which significantly improves assembly efficiency and accuracy. A welding sequence optimization model that takes into account the stress characteristics of key force-bearing structures is established to optimize the welding sequence of welds, effectively control welding deformation, and ensure the overall performance of the welded structure. Phased array ultrasonic imaging and machine vision technology are integrated to realize multi-information fusion intelligent evaluation of welding quality defects, build a quantitative defect evaluation system, and predict the defect expansion evolution behavior. A complete set of open press bed intelligent assembly and welding manufacturing methods and systems are formed to realize the digitization, automation and intelligence of the production process. It provides new ideas and methods for the intelligent manufacturing of large and complex welded structures, which is of great significance to promoting the application of intelligent manufacturing technology in the field of high-end equipment manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the process of the intelligent assembly and welding manufacturing method of the open press machine bed according to the embodiment of the present invention;

[0059] Figure 2 It is a structural schematic diagram of an intelligent assembly and welding manufacturing system for an open press machine bed according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0062] Figure 1 FIG. 1 is a flow chart of an intelligent assembly and welding manufacturing method for an open press machine bed according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0063] S101. According to the structural characteristics of the open press bed, the machine bed is divided into several independent functional modules, and an independent assembly process is formulated for each functional module, including sub-module division, positioning and clamping, welding stack and precision measurement. The machine bed is placed on a pre-set intelligent fixture, and the actual spatial position data of the machine bed is scanned by the displacement sensing unit of the intelligent fixture. The scanned actual spatial position data is compared with the design model of the machine bed, and the adjustment amount of the support point is calculated. According to the adjustment amount, each functional module is positioned and clamped to obtain an assembled open press bed.

[0064] The design of the assembly process flow of the open press bed is the key to achieving efficient and high-quality assembly. This section proposes a design method for the assembly process flow based on the modularization of the machine tool structure. First, the bed is divided into several independent functional modules according to the structural characteristics of the machine tool, such as the front wall, rear wall, left wall, right wall, base, etc. Then, an independent assembly process is formulated for each functional module, and the assembly relationship between the modules is comprehensively considered to reasonably arrange the overall assembly sequence.

[0065] In the design of module assembly process, taking the front wall module as an example, the strategy of "partition assembly + overall leveling" is adopted. Specifically, the front wall is first divided into two left and right sub-modules, which are assembled separately. When assembling, the sub-modules are first placed flat on the tooling platform, and the step pins and hydraulic clamps are used to achieve the initial positioning and clamping of the parts.

[0066] CO2 gas shielded welding is used to build up the submodules. During the welding process, infrared temperature measurement and visual tracking technology are used to monitor the welding temperature field and weld geometry in real time, and combined with the welding deformation prediction model, the welding parameters are adjusted online to control the welding deformation.

[0067] After the submodule is welded, it is accurately measured using a large-span digital measuring instrument, and mechanical processing is performed based on the measurement results to eliminate the geometric errors caused by welding deformation.

[0068] The processed left and right submodules are spliced ​​on a special tooling, and the laser tracker is used to dynamically track and adjust the posture to achieve high-precision docking. After the docking is completed, the welding deformation measurement points are arranged on the overall front wall module and measured using a three-coordinate measuring machine. According to the measurement results, the hydraulic leveling device integrated in the tooling is used to level the front wall as a whole to eliminate assembly errors.

[0069] Through the above modular assembly process design, the assembly efficiency can be greatly improved while ensuring the assembly accuracy. This method is not only applicable to the front wall module, but also to other functional modules. In actual production, a standardized and reusable module assembly process library can be formed, thereby realizing the rapid design and optimization of the open press bed assembly process.

[0070] The intelligent fixture consists of a positioning reference unit, a clamping execution unit, a displacement sensing unit and a control unit. The positioning reference unit is made of high-strength and high-rigidity marble material, and an array of adjustable support points are arranged on the surface for precise positioning of parts; the clamping execution unit uses a servo electric cylinder to achieve precise control of the clamping force; the displacement sensing unit is composed of a high-resolution laser displacement sensor for real-time monitoring of the clamping state; the control unit uses a bus-based distributed control system to achieve multi-channel servo control and data acquisition.

[0071] In an optional embodiment,

[0072] The displacement sensing unit of the intelligent fixture scans the actual spatial position data of the machine tool bed, compares the scanned actual spatial position data with the design model of the machine tool bed, calculates the adjustment amount of the support point, and positions and clamps each functional module according to the adjustment amount, so that the assembled open press bed includes:

[0073] Use a large-size three-dimensional coordinate measuring machine to measure the geometric accuracy of the machine tool bed and obtain the point cloud data of the machine tool bed that reflects the actual processing status of the bed;

[0074] Import the point cloud data of the machine tool bed into the computer, generate the actual processing model of the machine tool bed through the three-dimensional reconstruction technology, compare the actual processing model of the machine tool bed with the design model of the machine tool bed, extract and quantify the geometric deviation between the actual processing model of the machine tool bed and the design model of the machine tool bed, and establish the geometric deviation vector field between the actual processing model of the machine tool bed and the design model of the machine tool bed;

[0075] Based on the geometric deviation vector field of the actual processing model of the machine tool bed and the design model of the machine tool bed, a nonlinear mapping model of the assembly error between the machine tool bed assembly error and the machine tool bed adjustment parameters is constructed, where the machine tool bed adjustment parameters include the posture parameters of each functional module of the machine tool bed and the support parameters of the whole machine bed, and a neural network algorithm is used to train the nonlinear mapping model of the assembly error.

[0076] Obtain the target accuracy index and input the trained assembly error nonlinear mapping model, solve the optimal value of the machine tool bed adjustment parameter through the intelligent optimization algorithm, and generate the optimal value solution of the machine tool bed adjustment parameter;

[0077] By using the high-precision servo drive unit integrated in the intelligent fixture, the optimal value scheme of the machine tool bed adjustment parameters is converted into actual adjustment actions, and the functional modules of the machine tool bed are positioned and clamped. It is judged again whether the geometric accuracy indicators of the machine tool bed after adjustment reach the target accuracy. If not, the steps of obtaining the optimal value scheme of the bed adjustment parameters and converting it into actual adjustment actions are repeated iteratively until the geometric accuracy indicators of the machine tool bed after optimization and adjustment reach the target accuracy requirements, and the assembled open press bed is obtained.

[0078] For example, first, a large-scale three-coordinate measuring instrument is used to perform high-density measurement of various geometric accuracy indicators of the machine tool bed to obtain point cloud data of the machine tool bed that reflects the actual processing status of the bed. The three-coordinate measuring instrument is a high-precision coordinate measuring device that can accurately measure the size, shape, position, etc. of the workpiece. By arranging a large number of measuring points on the surface of the machine tool bed and collecting the three-dimensional coordinates of each measuring point, a point cloud data set reflecting the processing status of the bed can be obtained.

[0079] Then, the collected point cloud data of the machine tool bed is imported into the computer, and the actual processing model of the machine tool bed is generated using 3D reconstruction technology. 3D reconstruction refers to the process of restoring the complete geometric shape of an object through discrete data such as point clouds. Commonly used reconstruction algorithms include Poisson reconstruction and Delaunay triangulation. Through 3D reconstruction, a CAD model of the machine tool bed that is highly consistent with the actual object can be obtained. The model is compared with the design model, the geometric deviation between the two is extracted and quantified, and a deviation vector field is established. The geometric deviation reflects the degree of distortion of the actual processing model to the design model, which can be characterized by a spatial vector, whose direction indicates the direction of the deviation and whose size indicates the value of the deviation. For the entire machine tool bed, the geometric deviation has different characteristics at different positions, so a deviation vector field can be established.

[0080] Based on the obtained geometric deviation vector field of the actual processing model and the design model of the machine tool bed, a nonlinear mapping model of the assembly error between the machine tool bed assembly error and the adjustment parameters is constructed. This model reveals the intrinsic connection between the assembly error and the adjustment parameters, and is the mathematical basis for achieving assembly precision control. The assembly error of the machine tool bed refers to the degree of deviation between the actual posture and the ideal posture after the assembly of each functional module, which can be characterized by the Homogenous Transformation Matrix (HTM). The adjustment parameters refer to the posture parameters (such as position and angle) of each functional module during installation and the support parameters of the whole machine (such as the coordinates of the support point). Due to the complexity of the machine tool bed, the assembly error and the adjustment parameters are highly nonlinear, and it is difficult to accurately describe them with an analytical model. To this end, this paper uses a powerful nonlinear fitting tool, neural network, to construct the mapping relationship between the two. By selecting typical historical data to train the neural network, it can make intelligent predictions on the mapping law between unknown adjustment parameters and assembly errors.

[0081] After obtaining the trained nonlinear mapping model of assembly error, the target accuracy index is input into the model, and the optimal value of the adjustment parameter is solved with the help of the intelligent optimization algorithm, thereby generating a set of optimal bed adjustment solutions. The so-called target accuracy index refers to the design value or allowable deviation of various geometric accuracy of the machine tool bed, such as flatness, verticality, etc. These indicators are input into the mapping model, and the model can predict a set of optimal adjustment parameters corresponding to the accuracy requirements. The solution process here can be based on classical optimization theory, using intelligent optimization technologies such as genetic algorithms and particle swarm algorithms to search for the combination of adjustment parameters with the smallest assembly error in the parameter space, quickly lock the global optimal solution, and generate an adjustment plan. The plan includes the optimized values ​​of the installation posture of each functional module and the support status of the whole machine.

[0082] Finally, the optimal value scheme of the machine bed adjustment parameters is converted into an actual executable adjustment action, so as to optimize and adjust the posture of each functional module of the machine bed and the support status of the whole machine, and complete a closed-loop iteration. In this process, this paper innovatively introduces intelligent fixtures into machine tool assembly. The intelligent fixture is integrated with a high-precision servo drive unit, which can control the positioning and clamping mechanism in real time according to the optimized adjustment scheme, automatically guide the functional module to move to the optimal posture, and apply the optimal clamping force, thereby converting the digital adjustment scheme into an actual assembly action. After the adjustment is completed, the measurement steps are repeated to detect the geometric accuracy of the optimized machine bed, and determine whether the target accuracy requirements are met. If not, the above measurement-adjustment-remeasurement process is repeated and iterated, and the optimization rules are continuously extracted from data-driven, and the assembly accuracy is continuously improved until the expected goal is achieved. Through the closed-loop optimization mechanism of dynamic association between measurement data and adjustment schemes, the assembly accuracy of the machine bed can be continuously improved in multiple iterations, and finally reliable control of assembly quality can be achieved.

[0083] In this embodiment, the bed point cloud data of the actual processing state is obtained by three-coordinate measurement, and the geometric deviation is solved by comparing with the design model, so that the processing error of each component can be accurately quantified. The optimal value of the adjustment parameter is solved by the intelligent optimization algorithm, and then the intelligent fixture performs precise adjustment, so as to effectively compensate for the processing error and significantly improve the final assembly accuracy. The traditional manual experience assembly method is transformed into an automated assembly process based on measured data, digital models and intelligent algorithms. Advanced algorithms such as neural networks are used for modeling and optimization to reduce human intervention and improve assembly efficiency and consistency. The automated assembly process greatly saves manpower and reduces labor intensity. At the same time, through digital modeling and intelligent optimization, traditional repeated trial assembly and debugging are avoided, thereby greatly shortening the entire assembly cycle. Due to the improvement of assembly accuracy, the error compensation amount during machine tool operation is reduced, the movement is smoother, and the wear of components is reduced, thereby improving the reliability and service life of the whole machine. While achieving precise assembly, it also greatly improves production efficiency, product quality and flexibility.

[0084] In an optional embodiment,

[0085] Obtain the target accuracy index and input the trained assembly error nonlinear mapping model. Use the intelligent optimization algorithm to solve the optimal value of the machine tool bed adjustment parameters. The optimal value solution for the machine tool bed adjustment parameters includes:

[0086] The core concepts, attributes and semantic relationships in the machine tool field are defined through ontology modeling methods. Natural language processing technology is used to extract machine tool accuracy-related entities, attributes and relationship facts from structured and unstructured data sources. Knowledge fusion technology is used to perform representation learning, disambiguation links and consistency reasoning on multi-source heterogeneous machine tool accuracy knowledge, and a machine tool accuracy knowledge graph is constructed.

[0087] The conceptual nodes and semantic relationships related to causal analysis are extracted from the constructed machine tool accuracy knowledge graph to form causal prior knowledge, introduce domain scope constraints, and construct a machine tool error causal graph through knowledge-guided causal structure learning strategy;

[0088] The error data in the machine tool bed assembly process is obtained to construct training samples. The machine tool error cause-effect diagram is used as prior knowledge. The nonlinear mapping model of the machine tool assembly error is trained through a deep learning algorithm. The nonlinear mapping relationship between the machine tool influencing factors and the assembly error is established. The target accuracy index is input into the trained nonlinear mapping model of the machine tool assembly error. The optimal value of the machine tool bed adjustment parameters is solved through an intelligent optimization algorithm, and the optimal value solution of the machine tool bed adjustment parameters is generated.

[0089] After generating the optimal value solution for machine tool bed adjustment parameters, it also includes:

[0090] The machine tool bed is adjusted according to the optimal value plan of the machine tool bed adjustment parameters, the adjusted machine tool error is tested, the adjusted machine tool error test data is collected, the feature extraction and fusion of multi-source heterogeneous error indicators in the machine tool error test data are performed to obtain the machine tool error pattern vector, and the obtained machine tool error pattern vector is input into the machine tool error causal graph model. By performing a reverse search in the machine tool error causal graph, a causal path from the starting factor to the target error is generated, and the comprehensive influence intensity of the causal path is calculated. The key factors that have the greatest impact on the target error are identified through the factor sorting strategy, the error propagation chain schematic diagram of the key factors is extracted, and an error diagnosis report is automatically generated.

[0091] Exemplarily, the ontology modeling method is first used to define the core concepts, attributes and semantic relationships in the field of machine tools, and to construct a machine tool precision ontology model. The core concepts include machine tool type, error type, error source, measurement method, etc., the attributes include error value, influencing factor, compensation parameter, etc., and the semantic relationships include causal relationship, composition relationship, correlation relationship, etc. Then, natural language processing techniques such as named entity recognition and relationship extraction are used to extract entities, attributes and relationship facts related to machine tool precision from structured data (such as databases, knowledge bases, etc.) and unstructured data (such as literature, reports, etc.). Next, knowledge fusion techniques such as representation learning, disambiguation linking, and consistency reasoning are used to fuse and reason multi-source heterogeneous machine tool precision knowledge and construct a high-quality machine tool precision knowledge graph. Representation learning refers to mapping entities and relationships to low-dimensional dense vector space and learning their semantic representations; disambiguation linking refers to identifying mentions referring to the same entity or relationship in different data sources and linking them to unique nodes in the knowledge graph; consistency reasoning refers to using the constraint rules defined by the ontology to detect and repair errors in the knowledge graph to ensure the consistency and accuracy of knowledge.

[0092] Concept nodes (such as error types, influencing factors, etc.) and semantic relationships (such as causal relationships, correlation relationships, etc.) related to causal analysis are extracted from the machine tool accuracy knowledge graph to form prior knowledge for causal analysis of machine tool errors. Then, causal scope constraints defined by domain experts, such as time range and spatial range, are introduced to reduce the interference of irrelevant concepts and relationships. Next, knowledge-guided causal structure learning strategies, such as constraint-based structure learning and score-based structure learning, are used to learn the causal dependency of machine tool errors from prior knowledge and observation data, and to construct a causal graph model for machine tool errors. Causal structure learning refers to searching for the optimal causal graph structure under given observation data and prior knowledge. Commonly used methods include constraint learning based on independence test and structure learning based on Bayesian score.

[0093] Furthermore, the error data in the machine tool bed assembly process, such as the influencing factor parameters (such as temperature, vibration, etc.) and error measurement values ​​(such as positioning accuracy, repeated positioning accuracy, etc.), are obtained to construct the training samples of the machine tool assembly error. Then, the machine tool error causal graph is used as the prior knowledge, and the deep learning algorithm, such as feedforward neural network, convolutional neural network, etc., is used to train the nonlinear mapping model of the machine tool assembly error. The model takes the influencing factor parameters as input and the error measurement values ​​as output, and can establish the nonlinear mapping relationship between the machine tool influencing factors and the assembly error. The deep learning algorithm can fit complex nonlinear functions through multi-layer nonlinear transformations and has powerful feature learning and representation capabilities. The prior knowledge provided by the machine tool error causal graph, such as the dependency between the influencing factors, is used to guide the parameter optimization and structural design of the deep learning model, thereby improving the training efficiency and generalization ability of the model.

[0094] The target accuracy index (such as the target positioning accuracy value) is input into the trained nonlinear mapping model of machine tool assembly error, and the optimal value of the machine tool bed adjustment parameters is solved through intelligent optimization algorithms, such as genetic algorithm, particle swarm algorithm, etc. Intelligent optimization algorithms search the parameter space by simulating biological evolution or group behavior to find the parameter combination that minimizes the objective function (i.e., assembly error). In this paper, the nonlinear mapping model of machine tool assembly error is used as the fitness function of the optimization algorithm to evaluate the advantages and disadvantages of different adjustment parameter combinations. At the same time, the constraint information provided by the machine tool error causal graph is introduced, such as the parameter value range and the correlation between parameters, to reduce the optimization search space and improve the optimization efficiency and interpretability. Finally, the optimal value scheme of the machine tool bed adjustment parameters is generated to guide the machine tool adjustment process.

[0095] After generating the optimal value scheme for the machine tool bed adjustment parameters, the machine tool bed is adjusted according to the scheme, and the adjusted machine tool error is tested, and the adjusted machine tool error test data is collected. Then, feature engineering techniques such as feature extraction, feature selection, and feature fusion are used to process the multi-source heterogeneous error indicators in the machine tool error test data to obtain the machine tool error pattern vector. Feature extraction refers to extracting effective feature representations from the original error data, such as time domain features, frequency domain features, etc.; feature selection refers to selecting the most discriminative feature subset from the feature set, such as feature selection based on statistical tests, feature selection based on machine learning, etc.; feature fusion refers to combining and transforming multiple features to generate higher-level feature representations, such as principal component analysis and independent component analysis. Then, the machine tool error pattern vector is input into the machine tool error causal graph model, and a causal path from the target error node to the starting factor node is generated by reverse searching in the causal graph. Each causal path represents a possible error propagation chain, in which the node represents the error state or influencing factor, and the edge represents the causal dependency relationship. Then, the comprehensive impact intensity of each causal path is calculated, and the path length, causal strength, prior probability and other factors are comprehensively considered to quantify the impact of the path on the target error. Finally, the factor sorting strategy, such as sorting based on impact intensity and sorting based on sensitivity, is adopted to identify the key factors that have the greatest impact on the target error, extract the error propagation chain diagram of the key factors, and automatically generate an error diagnosis report to guide the analysis and improvement of machine tool errors.

[0096] In this embodiment, a knowledge base in the field of machine tool precision is constructed through ontology modeling and knowledge extraction, which provides rich prior knowledge for causal analysis; the machine tool error causal graph characterizes the generation mechanism and propagation law of machine tool errors through causal structure learning, and provides a causal basis for intelligent optimization decision-making; the machine tool assembly error nonlinear mapping model establishes a nonlinear relationship between influencing factors and assembly errors through a deep learning algorithm, which provides strong data support for error prediction; the optimal value of the machine tool bed adjustment parameter is generated through an intelligent optimization algorithm, which provides a precise adjustment strategy for machine tool error control; the machine tool error diagnosis reveals the root cause of the error through causal reasoning and key factor analysis, which provides powerful guidance for improving machine tool precision.

[0097] S102. A welding sequence optimization model for welds is established for the key force-bearing structure of the open press bed, with the maximum displacement deformation in the characteristic direction of the key force-bearing structure as the optimization target, and the geometric constraints and process constraints of the welds as constraint conditions. The optimal welding sequence of the welds is obtained by solving the welding sequence optimization model.

[0098] In an optional embodiment,

[0099] Aiming at the key force-bearing structure of the open press bed, a welding sequence optimization model of the weld is established. The maximum displacement deformation in the characteristic direction of the key force-bearing structure is taken as the optimization target, and the geometric constraints and process constraints of the weld are taken as the constraint conditions. By solving the welding sequence optimization model, the optimal welding sequence of the weld is obtained, including:

[0100] For the key force-bearing structure of the open press bed, the 3D CAD model of the key force-bearing structure is imported, the geometric topological relationship of the weld on the key force-bearing structure is identified, and the finite element analysis software is used to perform thermal-mechanical coupling simulation of the welding process under different welding sequences, predict the post-welding deformation, and post-process the simulation and prediction results. The maximum displacement deformation in the characteristic direction of the key force-bearing structure is extracted as the objective function of the welding sequence optimization model. The geometric constraints and process constraints of the weld are used as constraints, and the initial welding sequence of the weld is obtained by solving the welding sequence optimization model;

[0101] In the process of welding the key stress-bearing structure according to the initial welding sequence, the welding state information is collected in real time through multi-sensor fusion, and the welding state information includes welding parameters, weld geometry, temperature field and strain field;

[0102] Preprocess the collected welding state information, reconstruct the preprocessed welding state information into a state space vector, extract key features reflecting the degree and type of deviation from the state space vector through signal processing methods, construct a deviation discriminant function, calculate the distance between the current welding state and the expected welding state according to the deviation discriminant function, and quantitatively compare it with the set deviation threshold, and judge whether a deviation occurs according to the comparison result;

[0103] If a deviation is detected, the real-time re-optimization operation of the welding sequence is initiated, and the current welding state is evaluated according to the deviation. According to the evaluation result of the current welding state, the constraints in the welding sequence optimization model are dynamically adjusted, and the heuristic search algorithm is used to quickly search for the optimal welding sequence under the new constraints. The difference between the adjusted welding sequence and the original welding sequence is quantitatively evaluated, the adjustment range of the welding sequence and the optimization effect are weighed, and finally the optimal welding sequence of the weld is output.

[0104] For example, for the key force-bearing structure of the open press bed, first import its three-dimensional CAD model to identify the geometric topological relationship of all welds on the key force-bearing structure. Then call the finite element analysis software to perform thermal-mechanical coupling simulation of the welding process under different welding sequences, and predict the post-welding deformation caused by different welding sequences through simulation. The simulation prediction results are post-processed and analyzed to extract the maximum displacement deformation of the key force-bearing structure in the characteristic direction, which is used as the objective function for judging the quality of the welding sequence. At the same time, the geometric constraints of the weld (such as the position, length, groove form, etc. of the weld) and the process constraints (such as welding accessibility, welding parameter range, etc.) are comprehensively considered, and they are used as constraints to construct a mathematical model for welding sequence optimization. By solving the optimization model, a set of weld welding sequences that meet the constraints and minimize the objective function can be obtained, which is used as the initial welding sequence.

[0105] When welding the key stress-bearing structure according to the initial welding sequence, multi-sensor fusion technology is used to collect information reflecting the welding status in real time. These status information include: welding power supply parameters (such as welding current, voltage, speed, etc.), weld geometry parameters (such as weld width, height, angle, etc.), weld temperature field distribution and strain field distribution. Different types of sensors are required to collect different status information, such as arc sensors to monitor welding power supply parameters, laser vision sensors to monitor weld geometry parameters, infrared thermal imagers to monitor weld temperature fields, and fiber Bragg grating sensors to monitor weld strain fields.

[0106] The collected raw state information data needs to be preprocessed, including removing noise interference, smoothing data jitter, and normalizing, so as to highlight the characteristic components in the data that reflect welding deviation. The preprocessed multi-source heterogeneous state information is reconstructed into a state space vector in a certain way (such as according to time series). Then, a series of key feature quantities (such as energy, frequency, amplitude, etc.) that can reflect the degree and type of deviation are extracted from the state space vector using signal processing methods (such as wavelet transform, empirical mode decomposition, etc.). Based on the feature quantity, a deviation discriminant function is constructed to measure the difference distance between the current actual welding state and the expected standard welding state. The difference distance is compared with the pre-set deviation threshold. If it exceeds the threshold, it is judged that a significant deviation has occurred and timely countermeasures need to be taken.

[0107] Once the welding deviation is detected, the real-time re-optimization of the welding sequence is immediately started. First, the current welding state containing the deviation is evaluated to determine the severity of the deviation and the possible impact on subsequent welding. Then, the relevant constraints in the original welding sequence optimization model are adjusted in a targeted manner to adapt to the current actual deviation state. The optimal welding sequence is re-searched under the new constraints to minimize and eliminate the adverse effects of the deviation. Considering the high requirements of real-time optimization on computational efficiency, the fruit fly algorithm, a heuristic search strategy, is used to quickly find the optimal solution. In order to balance the optimization effect and optimization amplitude, the difference between the new and old welding sequences is evaluated, and the welding sequence with moderate difference and best optimization effect is selected as the update solution. Finally, the welding sequence is dynamically switched to the updated optimal solution to guide the subsequent welding process. At the same time, the idea of ​​dynamic programming is borrowed in the optimization process, and the dynamic constraints involved in the optimization process are decomposed into a series of static sub-constraints to reduce the complexity of the optimization problem.

[0108] In an optional embodiment,

[0109] Extracting the maximum displacement deformation in the key stress-bearing structural characteristic direction as the objective function of the welding sequence optimization model includes:

[0110] The deformation sensitivity coefficient and importance weight are introduced, and the objective function is constructed according to the maximum displacement deformation, root mean square displacement deformation and maximum displacement deformation difference. The formula is as follows:

[0111] ;

[0112] Where x represents the welding sequence vector, n represents the total number of welds, and w i represents the importance weight of the ith weld, s i represents the deformation sensitivity coefficient of the ith weld, λ1 represents the balance coefficient of the maximum displacement deformation, Ω i represents the node set in the area affected by the ith weld, u j (x) represents the displacement deformation of the jth node under the weld sequence x, λ2 represents the balance coefficient of the root mean square displacement deformation, λ3 represents the balance coefficient of the maximum displacement deformation difference, u k (x) represents the displacement deformation of the kth node under the weld sequence x;

[0113] The formula for calculating the distance between the current welding state and the expected welding state based on the deviation discriminant function is as follows:

[0114] ;

[0115] Where D(·) represents the deviation discriminant function, y represents the current welding state feature vector, represents the expected welding state feature vector,α represents the weight coefficient of distance, S represents the covariance matrix of feature vector, and β represents the weight coefficient of similarity.

[0116] In this embodiment, by optimizing the welding sequence through finite element simulation, the post-welding deformation can be minimized, thereby improving the overall geometric accuracy of the welded parts. At the same time, combined with real-time monitoring and feedback adjustment, various deviations in the welding process are further suppressed to ensure the final welding quality. Through numerical simulation and intelligent optimization algorithms, the automation and intelligent calculation of the welding sequence are realized, reducing human intervention. The automatic optimization of the welding sequence can greatly shorten the process preparation cycle and improve manufacturing efficiency. Real-time monitoring and optimization adjustment also enable the production process to dynamically adapt to changes, thereby improving the flexibility of equipment manufacturing. The optimized welding sequence and parameters can reduce the amount of post-welding deformation, thereby reducing the cost of rework. At the same time, the thermal efficiency is higher, the welding time is shorter, and energy consumption can also be reduced. The improvement of welding quality directly improves the performance and reliability of key load-bearing structures, ensuring the structural integrity and safety of the equipment during service.

[0117] S103. Use the optimal welding sequence to weld the assembled open press bed. During the welding process, use phased array ultrasonic imaging technology to identify the weld type and defect characteristics, output the welding defect ultrasonic imaging results, and use machine vision algorithms to perform intelligent detection on the welding defect ultrasonic imaging results. According to the intelligent detection results, the expansion and evolution behavior of the defects is predicted to form a quantitative defect evaluation system, and realize multi-information fusion intelligent evaluation of welding quality.

[0118] Phased array ultrasound imaging uses a two-dimensional array phased array transducer to achieve dynamic focusing and scanning of ultrasound by controlling multiple independent array elements to obtain ultrasound imaging results with high resolution and high signal-to-noise ratio. The deviation discriminant function is a mathematical function used to quantitatively describe the degree of deviation between the current state and the expected ideal state.

[0119] In an optional embodiment,

[0120] The assembled open press bed is welded using the optimal welding sequence. During the welding process, phased array ultrasonic imaging technology is used to identify the weld type and defect characteristics, output the ultrasonic imaging results of welding defects, and use machine vision algorithms to perform intelligent detection of the ultrasonic imaging results of welding defects. The expansion and evolution behavior of defects is predicted based on the intelligent detection results, forming a quantitative defect evaluation system to achieve multi-information fusion intelligent evaluation of welding quality, including:

[0121] A two-dimensionally arranged phased array ultrasonic transducer is used to optimize the transmitting end sound field and the receiving end sound field through synthetic aperture focusing, spatial filtering and sparse array methods. Phase calibration and amplitude normalization measures are used to adjust the phase consistency problem and dynamic range problem in the sound field optimization process to obtain the target optimized sound field, and the phased array ultrasonic imaging image is obtained according to the target optimized sound field.

[0122] Based on the obtained phased array ultrasonic imaging images, cluster analysis algorithm and pattern recognition algorithm are used to extract imaging parameter optimization rules from historical phased array ultrasonic imaging images, and imaging parameters are adaptively optimized in combination with expert experience. The imaging parameters include imaging angle range, focusing criteria and imaging image size. After optimization, the best ultrasonic imaging image of weld defects is obtained;

[0123] Extract the amplitude parameters, phase parameters and spectrum parameters of the defect echo signal in the best ultrasonic imaging image, combine the acoustic field model and the defect scattering model, and invert the defect quantitative information corresponding to the best ultrasonic imaging image, including the type, size and orientation information of the defect; at the same time, for three-dimensional defects, reconstruct the three-dimensional defect morphology of the three-dimensional defect through a tomographic imaging method and a cross-section synthesis method;

[0124] The quantitative information of defects and the best ultrasonic imaging image corresponding to the reconstructed three-dimensional defect morphology are converted into a grayscale image, and the morphological and texture features of the defects in the grayscale image are extracted using a machine vision algorithm, and input into a pre-trained convolutional neural network for defect segmentation, recognition and positioning to obtain intelligent detection results of weld defects;

[0125] The defect information in the intelligent detection results is fused with the pre-acquired X-ray detection information, eddy current detection information and magnetic memory detection information. Combined with the material properties of the weld where the defect is located, the welding process parameters and the weld load, an association model between defects and weld structural integrity is established. The quantitative defect information in the intelligent detection results is input into the association model to predict the expansion and evolution behavior of the defects, form a quantitative evaluation system for defects, and finally realize multi-information fusion intelligent evaluation of welding quality.

[0126] For example, first, a two-dimensionally arranged phased array ultrasonic transducer is used as a sound wave transmitting and receiving device. The phased array transducer is composed of multiple independently controlled array elements. By adjusting the excitation delay time and amplitude weight of each array element, the dynamic focusing and scanning imaging of the ultrasound wave can be realized. In order to obtain high-quality imaging images, it is necessary to optimize the sound field at the transmitting end and the sound field at the receiving end. Commonly used sound field optimization methods include synthetic aperture focusing, spatial filtering, and sparse array design. Synthetic aperture focusing synthesizes a large-aperture high-resolution image by coherently superimposing multiple transmission and reception data; spatial filtering suppresses noise interference by weighted averaging the received data; sparse array design reduces the number of array elements and reduces system complexity by optimizing the array element layout. In the process of sound field optimization, the phase consistency problem and dynamic range problem between the array elements also need to be considered. The phase consistency problem refers to the phase deviation between the excitation signal and the received signal of each array element, which causes the focus point to deviate from the ideal position. The dynamic range problem refers to the large difference in the receiving amplitude of each array element, which leads to reduced image contrast. For the phase consistency problem, phase calibration measures can be adopted, that is, introducing a known reference signal during the transmission and reception process, measuring the phase offset of each array element relative to the reference signal, and compensating for it. For the dynamic range problem, amplitude normalization measures can be adopted, that is, normalizing the amplitude of the received signal of each array element to make its amplitude distribution more uniform. After the above sound field optimization and adjustment processing, a target optimized sound field can be obtained that suppresses noise interference and improves the sensitivity of defect detection, and then a high-resolution, high signal-to-noise ratio phased array ultrasonic imaging image can be obtained based on the target optimized sound field.

[0127] Secondly, in order to further improve the quality of defect imaging, it is necessary to adaptively optimize the phased array ultrasonic imaging parameters for different weld types and defect characteristics. First, based on the obtained phased array ultrasonic imaging images, the historical phased array ultrasonic imaging images are classified using a clustering analysis algorithm, and the imaging images with similar defect characteristics are clustered into one category. Then, a pattern recognition algorithm is used for each type of imaging image to extract the characteristic pattern of the defect echo signal and establish a standard feature template for this type of defect. In the actual detection process, after obtaining the phased array ultrasonic imaging image of the weld to be tested, it is matched with various defect standard feature templates to identify the defect category to which it belongs, and the historical imaging image corresponding to this type of defect is referred to to extract the verified optimal imaging parameter combination as the parameter optimization rule for the current weld defect imaging. The imaging parameters mainly include imaging angle range, focusing criterion, imaging image size, etc. The imaging angle range refers to the range of variation of the ultrasonic incident angle, which determines the size and shape of the imaging area; the focusing criterion refers to the position and intensity distribution of the focus point, which determines the concentration of the defect echo signal; the imaging image size refers to the resolution and field of view of the image, which determines the clarity of the defect details. The extracted imaging parameter optimization rules are combined with expert experience to adaptively optimize the current weld defect imaging parameters, and finally obtain the best ultrasonic imaging image that can display the defect detail characteristics to the greatest extent.

[0128] Then, the best ultrasonic imaging image is post-processed and defect information is extracted. First, the signal-to-noise ratio and clarity of the defect echo signal are improved by image enhancement, filtering and denoising. Then, the characteristic parameters of the defect echo signal, such as amplitude, phase, and spectrum, are extracted. The amplitude parameter reflects the size and type of the defect, the phase parameter reflects the position and orientation of the defect, and the spectrum parameter reflects the surface morphology and material properties of the defect. Based on the extracted defect characteristic parameters, combined with the propagation law of ultrasonic waves in the weld material, an acoustic field model is established to describe the interaction mechanism between ultrasonic waves and defects, and reveal the corresponding relationship between defect characteristic parameters and defect physical properties. At the same time, for common defect types, such as cracks, unfused, slag inclusions, etc., corresponding defect scattering models are established to simulate the scattering characteristics of different types of defects under the action of ultrasonic waves. The quantitative information of the defect corresponding to the best ultrasonic imaging image is inverted by combining the acoustic field model and the defect scattering model, including the type, size and orientation information of the defect. It should be noted that ultrasonic imaging can usually only obtain a two-dimensional cross-sectional image of the defect. For defects extending in three-dimensional space, a single imaging image often has incomplete information. Therefore, for three-dimensional defects, it is necessary to scan and image them multiple times from different angles to obtain a series of parallel or cross-sectional images of defects, and then use the tomography method to reconstruct the three-dimensional morphology of the defects through the reconstruction algorithm of the image sequence, such as the back-projection algorithm, the algebraic reconstruction algorithm, etc.; or use the cross-sectional synthesis method to connect the defect section contours through the interpolation fitting algorithm to synthesize the three-dimensional defect model. After the above quantitative information inversion and three-dimensional morphology reconstruction, a comprehensive and accurate defect characterization result can be obtained.

[0129] Next, the inverted defect quantitative information and the optimal ultrasonic imaging image corresponding to the reconstructed three-dimensional defect morphology are visualized and converted into grayscale images or pseudo-color images for subsequent intelligent analysis and discrimination. The converted images are preprocessed using image processing algorithms such as noise removal, edge enhancement, segmentation extraction, etc. to highlight the morphological contours and detailed textures of the defects. Then, machine vision algorithms such as Hough transform, region growing, template matching, etc. are used to extract morphological features such as geometric shape, topological structure, grayscale distribution, and texture features such as texture direction, texture frequency, and texture contrast of the defects in the image. The extracted morphological features and texture features are input into the pre-built and trained convolutional neural network model to achieve intelligent classification and recognition of defects. Convolutional neural network is a deep learning model that extracts local features of the image through the convolution layer, reduces the feature dimension through the pooling layer, and realizes feature classification through the fully connected layer. It has powerful feature learning and classification discrimination capabilities. A large number of historical defect image samples are input into the convolutional neural network for training. The network can autonomously learn the mapping relationship between defect features and defect types to form defect discrimination rules. By inputting the current defect image to be tested into the trained network, the type of defect in the image can be automatically identified, and the defect can be accurately located by analyzing the pixel coordinates of the defect area. At this point, the intelligent detection results of weld defects can be obtained, including key information such as defect type, defect location, and defect size.

[0130] Finally, in order to further improve the reliability of defect detection and evaluation, defect information complementary to the ultrasonic detection results can be obtained through X-ray detection, eddy current detection, magnetic memory detection and other methods. For example, X-ray detection has a higher sensitivity for volume defects, such as slag inclusions and pores, while ultrasonic detection has a higher resolution for planar defects, such as cracks and lack of fusion. Eddy current detection and magnetic memory detection have unique advantages for surface and near-surface defects, such as surface cracks and stress concentration. The defect information obtained by ultrasonic detection is fused with the defect information obtained by X-ray, eddy current, and magnetic memory detection. With the help of data association analysis, evidence theory and other methods, the defect characteristics are characterized three-dimensionally from the perspective of multiple physical fields, and the causes, morphologies, and severity of defects are fully characterized. On the basis of defect information fusion, it is also necessary to combine the material properties of the weld where the defect is located, welding process parameters, stress state and other factors to comprehensively consider the impact of defects on the structural integrity and service performance of the weld. Through finite element simulation analysis, fracture mechanics calculation and other methods, a quantitative correlation model between defects and weld structural performance is established. The quantitative defect parameters inverted from the ultrasonic intelligent detection results are input into the correlation model for evaluation and calculation, the expansion and evolution behavior of defects under actual loads is predicted, the degree of harm of defects is judged, and quantitative evaluation indicators are given for the safety and reliability of weld structures. Based on key parameters such as defect size, defect morphology, defect location, material fracture toughness, and stress intensity factor, a set of scientific and standardized defect quantitative evaluation systems and scrap standards are formed to achieve a leap from qualitative defect detection to quantitative defect evaluation. Finally, through multi-information fusion intelligent evaluation, the welding quality is comprehensively diagnosed and predicted, potential structural safety hazards are discovered and eliminated in a timely manner, and the reliable operation of the welded structure is guaranteed.

[0131] In this embodiment, through the acoustic field optimization of phased array imaging technology, adaptive imaging parameter selection and other measures, high-quality defect imaging results can be obtained, and the resolution and signal-to-noise ratio of defect detection can be improved. By integrating the information of different detection methods, the defect characteristics are comprehensively characterized from multiple angles and multiple levels, and the detection results are more accurate and reliable. With the help of artificial intelligence technologies such as machine vision algorithms and deep learning models, defects can be automatically and efficiently identified and located from imaging data, and intelligent and automated defect detection can be realized to reduce manual intervention. Not only can static parameters such as the type, size and location of defects be obtained, but also the dynamic expansion and evolution behavior of defects can be predicted through the correlation model between defects and structural integrity, so as to make a comprehensive quantitative evaluation of the degree of harm of defects. Based on the quantitative evaluation results of defects, welds with serious defects can be accurately repaired, and welding process parameters can be optimized through big data analysis, welding quality can be continuously improved, and the reliability of equipment use can be improved. The efficient and intelligent welding quality detection process is realized, which provides strong technical support for the refined control of weld quality and equipment safety assessment.

[0132] Figure 2 FIG. 1 is a schematic diagram of the structure of the intelligent assembly and welding manufacturing system for the open press bed according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0133] The first unit is used to divide the machine bed into several independent functional modules according to the structural characteristics of the open press bed, and formulate an independent assembly process for each functional module, including sub-module division, positioning and clamping, welding stack and precision measurement, and place the machine bed on a pre-set intelligent fixture, scan the actual spatial position data of the machine bed through the displacement sensing unit of the intelligent fixture, compare the scanned actual spatial position data with the design model of the machine bed, calculate the adjustment amount of the support point, and position and clamp each functional module according to the adjustment amount to obtain an assembled open press bed;

[0134] The second unit is used to establish a welding sequence optimization model for the key force-bearing structure of the open press bed, taking the maximum displacement deformation in the characteristic direction of the key force-bearing structure as the optimization target, taking the geometric constraints and process constraints of the weld as the constraint conditions, and obtaining the optimal welding sequence of the weld by solving the welding sequence optimization model;

[0135] The third unit is used to weld the assembled open press bed using the optimal welding sequence. During the welding process, phased array ultrasonic imaging technology is used to identify the weld type and defect characteristics, output the ultrasonic imaging results of welding defects, and use machine vision algorithms to perform intelligent detection on the ultrasonic imaging results of welding defects. The expansion and evolution behavior of defects is predicted based on the intelligent detection results, forming a quantitative defect evaluation system and realizing multi-information fusion intelligent evaluation of welding quality.

[0136] According to a third aspect of the embodiments of the present invention,

[0137] An electronic device is provided, comprising:

[0138] processor;

[0139] a memory for storing processor-executable instructions;

[0140] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0141] A fourth aspect of the embodiments of the present invention is:

[0142] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0143] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent assembly and welding manufacturing method for an open press bed, characterized in that: include: According to the structural characteristics of the open press bed, the machine bed is divided into several independent functional modules, and an independent assembly process is formulated for each functional module, including sub-module division, positioning and clamping, welding stacking and precision measurement. The machine bed is placed on a pre-set intelligent fixture, and the displacement sensing unit of the intelligent fixture is used to scan the actual spatial position data of the machine bed. The scanned actual spatial position data is compared with the design model of the machine bed, and the adjustment amount of the support point is calculated. According to the adjustment amount, each functional module is positioned and clamped to obtain an assembled open press bed; For the key force-bearing structure of the open press bed, the 3D CAD model of the key force-bearing structure is imported, the geometric topological relationship of the weld on the key force-bearing structure is identified, and the finite element analysis software is used to perform thermal-mechanical coupling simulation of the welding process under different welding sequences, predict the post-welding deformation, and post-process the simulation and prediction results. The maximum displacement deformation in the characteristic direction of the key force-bearing structure is extracted as the objective function of the welding sequence optimization model. The geometric constraints and process constraints of the weld are used as constraints, and the initial welding sequence of the weld is obtained by solving the welding sequence optimization model; In the process of welding the key stress-bearing structure according to the initial welding sequence, the welding state information is collected in real time through multi-sensor fusion, and the welding state information includes welding parameters, weld geometry, temperature field and strain field; Preprocess the collected welding state information, reconstruct the preprocessed welding state information into a state space vector, extract key features reflecting the degree and type of deviation from the state space vector through signal processing methods, construct a deviation discriminant function, calculate the distance between the current welding state and the expected welding state according to the deviation discriminant function, and quantitatively compare it with the set deviation threshold, and judge whether a deviation occurs according to the comparison result; If a deviation is detected, the real-time re-optimization operation of the welding sequence is started, and the current welding state is evaluated according to the deviation. According to the evaluation result of the current welding state, the constraints in the welding sequence optimization model are dynamically adjusted, and the heuristic search algorithm is used to quickly search for the optimal welding sequence under the new constraints. The difference between the adjusted welding sequence and the original welding sequence is quantitatively evaluated, and the adjustment range and optimization effect of the welding sequence are weighed. Finally, the optimal welding sequence of the weld is output; The assembled open press bed is welded using the optimal welding sequence. During the welding process, phased array ultrasonic imaging technology is used to identify the weld type and defect characteristics, and the ultrasonic imaging results of the welding defects are output. The machine vision algorithm is used to perform intelligent detection on the ultrasonic imaging results of the welding defects. The expansion and evolution behavior of the defects is predicted based on the intelligent detection results, forming a quantitative defect evaluation system and realizing multi-information fusion intelligent evaluation of welding quality.

2. The method according to claim 1, characterized in that The displacement sensing unit of the intelligent fixture scans the actual spatial position data of the machine tool bed, compares the scanned actual spatial position data with the design model of the machine tool bed, calculates the adjustment amount of the support point, and positions and clamps each functional module according to the adjustment amount, so that the assembled open press bed includes: Use a large-size three-dimensional coordinate measuring machine to measure the geometric accuracy of the machine tool bed and obtain the point cloud data of the machine tool bed that reflects the actual processing status of the bed; Import the point cloud data of the machine tool bed into the computer, generate the actual processing model of the machine tool bed through the three-dimensional reconstruction technology, compare the actual processing model of the machine tool bed with the design model of the machine tool bed, extract and quantify the geometric deviation between the actual processing model of the machine tool bed and the design model of the machine tool bed, and establish the geometric deviation vector field between the actual processing model of the machine tool bed and the design model of the machine tool bed; Based on the geometric deviation vector field of the actual processing model of the machine tool bed and the design model of the machine tool bed, a nonlinear mapping model of the assembly error between the machine tool bed assembly error and the machine tool bed adjustment parameters is constructed, where the machine tool bed adjustment parameters include the posture parameters of each functional module of the machine tool bed and the support parameters of the whole machine bed, and a neural network algorithm is used to train the nonlinear mapping model of the assembly error. Obtain the target accuracy index and input the trained assembly error nonlinear mapping model, solve the optimal value of the machine tool bed adjustment parameter through the intelligent optimization algorithm, and generate the optimal value solution of the machine tool bed adjustment parameter; By using the high-precision servo drive unit integrated in the intelligent fixture, the optimal value scheme of the machine tool bed adjustment parameters is converted into actual adjustment actions, and the functional modules of the machine tool bed are positioned and clamped. It is judged again whether the geometric accuracy indicators of the machine tool bed after adjustment reach the target accuracy. If not, the steps of obtaining the optimal value scheme of the bed adjustment parameters and converting it into actual adjustment actions are repeated iteratively until the geometric accuracy indicators of the machine tool bed after optimization and adjustment reach the target accuracy requirements, and the assembled open press bed is obtained.

3. The method according to claim 2, characterized in that Obtain the target accuracy index and input the trained assembly error nonlinear mapping model. Use the intelligent optimization algorithm to solve the optimal value of the machine tool bed adjustment parameters. The optimal value solution for the machine tool bed adjustment parameters includes: The core concepts, attributes and semantic relationships in the machine tool field are defined through ontology modeling methods. Natural language processing technology is used to extract machine tool accuracy-related entities, attributes and relationship facts from structured and unstructured data sources. Knowledge fusion technology is used to perform representation learning, disambiguation links and consistency reasoning on multi-source heterogeneous machine tool accuracy knowledge, and a machine tool accuracy knowledge graph is constructed. The conceptual nodes and semantic relationships related to causal analysis are extracted from the constructed machine tool accuracy knowledge graph to form causal prior knowledge, introduce domain scope constraints, and construct a machine tool error causal graph through knowledge-guided causal structure learning strategy; The error data in the machine tool bed assembly process is obtained to construct training samples. The machine tool error cause-effect diagram is used as prior knowledge. The nonlinear mapping model of the machine tool assembly error is trained through a deep learning algorithm. The nonlinear mapping relationship between the machine tool influencing factors and the assembly error is established. The target accuracy index is input into the trained nonlinear mapping model of the machine tool assembly error. The optimal value of the machine tool bed adjustment parameters is solved through an intelligent optimization algorithm, and the optimal value solution of the machine tool bed adjustment parameters is generated. After generating the optimal value solution for machine tool bed adjustment parameters, it also includes: The machine tool bed is adjusted according to the optimal value plan of the machine tool bed adjustment parameters, the adjusted machine tool error is tested, the adjusted machine tool error test data is collected, the feature extraction and fusion of multi-source heterogeneous error indicators in the machine tool error test data are performed to obtain the machine tool error pattern vector, and the obtained machine tool error pattern vector is input into the machine tool error causal graph model. By performing a reverse search in the machine tool error causal graph, a causal path from the starting factor to the target error is generated, and the comprehensive influence intensity of the causal path is calculated. The key factors that have the greatest impact on the target error are identified through the factor sorting strategy, the error propagation chain schematic diagram of the key factors is extracted, and an error diagnosis report is automatically generated.

4. The method according to claim 1, characterized in that: Extracting the maximum displacement deformation in the key stress-bearing structural characteristic direction as the objective function of the welding sequence optimization model includes: The deformation sensitivity coefficient and importance weight are introduced, and the objective function is constructed according to the maximum displacement deformation, root mean square displacement deformation and maximum displacement deformation difference. The formula is as follows: ； Where x represents the welding sequence vector, n represents the total number of welds, and w i represents the importance weight of the ith weld, s i represents the deformation sensitivity coefficient of the ith weld, λ1 represents the balance coefficient of the maximum displacement deformation, Ω i represents the node set in the area affected by the ith weld, u j (x) represents the displacement deformation of the jth node under the weld sequence x, λ2 represents the balance coefficient of the root mean square displacement deformation, λ3 represents the balance coefficient of the maximum displacement deformation difference, u k (x) represents the displacement deformation of the kth node under the weld sequence x; The formula for calculating the distance between the current welding state and the expected welding state based on the deviation discriminant function is as follows: ； Where D(·) represents the deviation discriminant function, y represents the current welding state feature vector, represents the expected welding state feature vector, α represents the weight coefficient of distance, S represents the covariance matrix of feature vector, and β represents the weight coefficient of similarity.

5. The method according to claim 1, characterized in that The assembled open press bed is welded using the optimal welding sequence. During the welding process, phased array ultrasonic imaging technology is used to identify the weld type and defect characteristics, output the ultrasonic imaging results of welding defects, and use machine vision algorithms to perform intelligent detection of the ultrasonic imaging results of welding defects. The expansion and evolution behavior of defects is predicted based on the intelligent detection results, forming a quantitative defect evaluation system to achieve multi-information fusion intelligent evaluation of welding quality, including: A two-dimensionally arranged phased array ultrasonic transducer is used to optimize the transmitting end sound field and the receiving end sound field through synthetic aperture focusing, spatial filtering and sparse array methods. Phase calibration and amplitude normalization measures are used to adjust the phase consistency and dynamic range problems in the sound field optimization process to obtain the target optimized sound field, and the phased array ultrasonic imaging image is obtained according to the target optimized sound field. Based on the obtained phased array ultrasonic imaging images, cluster analysis algorithm and pattern recognition algorithm are used to extract imaging parameter optimization rules from historical phased array ultrasonic imaging images, and imaging parameters are adaptively optimized in combination with expert experience. The imaging parameters include imaging angle range, focusing criteria and imaging image size. After optimization, the best ultrasonic imaging image of weld defects is obtained; Extract the amplitude parameters, phase parameters and spectrum parameters of the defect echo signal in the best ultrasonic imaging image, combine the acoustic field model and the defect scattering model, and invert the defect quantitative information corresponding to the best ultrasonic imaging image, including the type, size and orientation information of the defect; at the same time, for three-dimensional defects, reconstruct the three-dimensional defect morphology of the three-dimensional defect through a tomographic imaging method and a cross-section synthesis method; The quantitative information of defects and the best ultrasonic imaging image corresponding to the reconstructed three-dimensional defect morphology are converted into a grayscale image, and the morphological and texture features of the defects in the grayscale image are extracted using a machine vision algorithm, and input into a pre-trained convolutional neural network for defect segmentation, recognition and positioning to obtain intelligent detection results of weld defects; The defect information in the intelligent detection results is fused with the pre-acquired X-ray detection information, eddy current detection information and magnetic memory detection information. Combined with the material properties of the weld where the defect is located, the welding process parameters and the weld load, an association model between defects and weld structural integrity is established. The quantitative defect information in the intelligent detection results is input into the association model to predict the expansion and evolution behavior of the defects, form a quantitative evaluation system for defects, and finally realize multi-information fusion intelligent evaluation of welding quality.

6. An intelligent assembly and welding manufacturing system for an open press bed, used to implement the method described in any one of claims 1 to 5, characterized in that: include: The first unit is used to divide the machine bed into several independent functional modules according to the structural characteristics of the open press bed, and formulate an independent assembly process for each functional module, including sub-module division, positioning and clamping, welding stack and precision measurement, and place the machine bed on a pre-set intelligent fixture, scan the actual spatial position data of the machine bed through the displacement sensing unit of the intelligent fixture, compare the scanned actual spatial position data with the design model of the machine bed, calculate the adjustment amount of the support point, and position and clamp each functional module according to the adjustment amount to obtain an assembled open press bed; The second unit is used to establish a welding sequence optimization model for the key force-bearing structure of the open press bed, taking the maximum displacement deformation in the characteristic direction of the key force-bearing structure as the optimization target, taking the geometric constraints and process constraints of the weld as the constraint conditions, and obtaining the optimal welding sequence of the weld by solving the welding sequence optimization model; The third unit is used to weld the assembled open press bed using the optimal welding sequence. During the welding process, phased array ultrasonic imaging technology is used to identify the weld type and defect characteristics, output the ultrasonic imaging results of welding defects, and use machine vision algorithms to perform intelligent detection on the ultrasonic imaging results of welding defects. The expansion and evolution behavior of defects is predicted based on the intelligent detection results, forming a quantitative defect evaluation system and realizing multi-information fusion intelligent evaluation of welding quality.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images