Arc additive and subtractive forming defect diagnosis and intelligent control method and system based on data fusion

Abstract

The present invention belongs to the cross-technical field of computer-aided design, additive manufacturing and subtractive manufacturing, and is particularly a method and system for diagnosing and intelligently controlling arc additive and subtractive forming defects based on data fusion; obtaining multi-modal data signals in arc additive and subtractive manufacturing; extracting feature parameters in real time from the acquired multi-modal signals; constructing an SVM classification model, that is, obtaining a comprehensive feature vector by data fusion of feature parameters; training the SVM classification model; combining the trained SVM classification model with a defect classification recognition set, identifying the forming defect classification and obtaining the recognition result; according to the defect classification recognition result, performing real-time subtraction on the defective part of the product, adjusting the manufacturing-related processing parameters, and re-adding and subtracting. Compared with the prior art, the defects generated in the additive process are monitored in real time, and the defects that affect the product quality are subtracted in real time; thereby ensuring the quality of the processed parts and improving the utilization rate of the materials.

Description

Technical Field

[0001] The present invention belongs to the cross-technical field of computer-aided design, additive manufacturing and subtractive manufacturing, and in particular to a method and system for diagnosing and intelligently controlling arc additive and subtractive forming defects based on data fusion. Background Art

[0002] In modern manufacturing, computer-aided design (CAD) technology has become a key tool in the product development and manufacturing process. CAD technology allows engineers and designers to use computer software for precise design, modeling, and simulation, thereby improving design efficiency and product quality. In the early stages of product development, CAD can help quickly generate three-dimensional models, conduct design verification, and evaluate the feasibility of the design through simulation analysis. This technology reduces the need for physical prototypes, saving time and costs. With the improvement of computing power and the advancement of CAD software, CAD technology is not only limited to traditional two-dimensional drawing, but also supports complex three-dimensional design and multi-physics simulation. The results of multi-physics simulation can be fed back to the CAD system to optimize design parameters. This closed-loop design and simulation process allows engineers to identify potential problems in the early stages of product processing and optimize through multiple iterations, thereby reducing the production cost of the product.

[0003] Arc additive and subtractive hybrid manufacturing is an advanced manufacturing method that combines additive manufacturing and subtractive manufacturing technologies. It uses arc welding and traditional machining methods to complete the manufacture of parts on the same platform. This hybrid manufacturing method can give full play to the advantages of both technologies and overcome the limitations of a single manufacturing method. The principle of this technology is to deposit metal materials layer by layer through arc additive manufacturing (such as WAAM), and then perform subtractive machining on the additively manufactured parts through CNC machining (such as CNC milling and turning) to achieve the required surface accuracy, or use subtractive machining to change the size of the parts during the machining process so that the required part shape can be continued through additive means in the future. It combines the rapid prototyping capability of additive manufacturing and the high-precision machining capability of subtractive manufacturing, improves manufacturing efficiency and reduces material waste, because additive manufacturing can maximize the use of raw materials, while subtractive manufacturing is refined into the final shape. It also has the advantages of strong flexibility and integrated manufacturing.

[0004] However, arc additive and subtractive hybrid manufacturing technology is also a multi-scale, multi-physics field coupling process that evolves over time, such as temperature field, stress / strain field, mass transport, metal microstructure evolution, etc. Compared with other additive manufacturing technologies, this method has a large heat input, resulting in weaker control over larger molten pools, more complex metallurgical processes, and difficulty in accurately controlling energy, which leads to a greater possibility of defects. At the same time, the continuous scanning of the heat source produces a non-steady-state thermal cycle process, resulting in internal stress and phase evolution. Residual stress may be accompanied by unacceptable deformation, delamination, and fatigue performance degradation. In addition, conditions such as uneven material composition, unstable real-time process parameters, and fluctuating external conditions are also prone to cause porosity, slag inclusions, insufficient fusion, humps, and other phenomena.

[0005] In the prior art, only one signal is monitored and data collected during the arc additive and subtractive manufacturing process, and defects are detected by analyzing the collected single signal only after the product is formed. Since defects are only detected after manufacturing is completed, flaws generated during the manufacturing process cannot be corrected immediately, which may cause the entire part to be scrapped, thereby wasting materials, time and production costs. In addition, during the entire manufacturing process, small defects in the early stages may be amplified or evolve into more serious problems during the material deposition process of subsequent levels. If these defects are not handled in a timely manner, they will eventually affect the overall performance of the part. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for arc additive and subtractive forming defect diagnosis and intelligent control based on data fusion, so as to solve the problem of real-time detection and diagnosis of defects in arc additive and subtractive manufacturing. The present invention collects various signals in the arc additive manufacturing process through multiple sensors and performs feature extraction, and then establishes and trains a SVM classification model to achieve real-time defect diagnosis in the processing process; combined with subtractive manufacturing to remove the defective parts of the processed parts, the processing parameters are adjusted in time according to the processing defects judged by the system, so as to ensure the quality of the processed parts and improve the utilization rate of materials.

[0007] The technical solution of the present invention: The present invention discloses a method for diagnosing and intelligently controlling arc-assisted material forming defects based on data fusion, which specifically includes the following steps:

[0008] S1. Use infrared thermal imagers, Hall sensors, acoustic sensors, spectrometers, and vibration sensors to obtain multi-modal data signals in arc additive and subtractive manufacturing in real time;

[0009] S2, extracting characteristic parameters in real time from the acquired molten pool signal, audible sound signal, arc voltage signal, vibration signal and arc spectrum signal;

[0010] S3, constructing an SVM classification model, that is, obtaining a comprehensive feature vector by performing data fusion on feature parameters;

[0011] S4. Divide the training set and the verification set according to the feature parameters extracted in real time, input the training set into the SVM classification model for training, and verify the effectiveness of the SVM classification model according to the product forming defects;

[0012] S5. Combining the trained SVM classification model with the defect classification and recognition set is used for arc additive and subtractive forming defect classification and recognition, and obtaining defect classification and recognition results in the arc additive and subtractive forming process;

[0013] S6. According to the defect classification and identification results of the SVM classification model, the defective parts of the product are reduced in real time through the cutting and milling system, and the relevant processing parameters of arc additive and subtractive manufacturing are adjusted in time using the model-free adaptive iterative learning control algorithm to re-add and subtract materials.

[0014] A data fusion-based arc additive and subtractive material forming defect diagnosis and intelligent control system, the system comprises an infrared thermal imager, a Hall sensor, an acoustic sensor, a spectrometer, a vibration sensor, a computer, and a milling system; multimodal information in arc additive and subtractive material manufacturing is acquired by the infrared thermal imager, the Hall sensor, the acoustic sensor, the spectrometer, and the vibration sensor, and the acquired information is transmitted to the computer, which performs relevant data processing, thereby diagnosing the forming defects in the additive and subtractive material manufacturing, removing the generated product defects by using the milling system, and adjusting the processing parameters in time by using a model-free adaptive iterative learning control algorithm, and re-adding and subtracting materials to obtain processed parts with good forming quality.

[0015] Compared with the prior art, the present invention has the following significant advantages:

[0016] (1) The present invention combines the additive process with the subtractive process, monitors the defects generated in the additive process in real time, and performs real-time subtraction of the defects that affect product quality.

[0017] (2) The present invention diagnoses processing defects and adjusts processing parameters in real time according to the diagnosis results to increase or decrease material.

[0018] (3) The present invention collects and analyzes multimodal information in the arc material addition and subtraction process, including molten pool information, acoustic information, electrical information, spectral information, and vibration information, and extracts and fuses the multimodal information, thereby improving the accuracy of defect diagnosis in the arc material addition and subtraction process.

[0019] (4) The present invention combines graph theory methods with information collected by sensors. Compared with the finite element method, it can greatly reduce the time required to predict temperature distribution while ensuring the accuracy of the thermal model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of a method for detecting defects in arc additive and subtractive manufacturing.

[0021] Figure 2 This is the structural block diagram of the SVM classification model. DETAILED DESCRIPTION

[0022] The following is combined with Figure 1-Figure 2 The present invention is further described with reference to the accompanying drawings and specific embodiments.

[0023] The present invention discloses a method for diagnosing and intelligently controlling arc-assisted material forming defects based on data fusion, which specifically includes the following steps:

[0024] S1. Use infrared thermal imagers, Hall sensors, acoustic sensors, spectrometers, and vibration sensors to obtain multi-modal data signals in arc additive and subtractive manufacturing in real time;

[0025] S2. Extract characteristic parameters in real time from the acquired molten pool signal, audible sound signal, arc voltage signal, vibration signal and arc spectrum signal. These characteristics include various statistics, such as average value, peak value, etc.;

[0026] (1) For the molten pool signal, a computational heat transfer method based on graph theory is used to predict the temperature distribution of the component. The average value and standard deviation of the temperature predicted by the graph theory method will be used as the characteristic parameters of the molten pool signal. The temperature thermal characteristics predicted by the graph theory are combined with the thermal characteristics obtained by using a thermal imager during the arc additive manufacturing process for defect diagnosis.

[0027] Graph theory thermal modeling is to solve the continuous heat diffusion equation, then The internal positions of the parts The temporal and spatial temperature distribution of The calculation formula is:

[0028] ;

[0029] In the formula, For time, is the density, is the specific heat capacity, is the thermal conductivity, is the three-dimensional coordinate value inside the part, and the right side of the equation is the time Location Heat input In arc additive manufacturing, heat input is the energy provided by the arc.

[0030] (2) For arc spectrum signals, in order to enhance the spectral characteristics and eliminate the influence of background signals, the spectral data collected by the spectrometer are subjected to continuous spectrum removal. Then, the spectral lines are selected and FastICA is applied to perform dimensionality reduction processing on the spectral lines. The mean signal is used to replace the original signal to eliminate pulse interference, and then the characteristic parameter set of the arc spectrum signal is obtained.

[0031] (3) For audible sound signals, in order to extract effective characteristic parameters of arc sound signals, it is necessary to first denoise the arc sound signals. The wavelet threshold denoising method decomposes the signal into different scales through wavelet decomposition, and controls the high-frequency coefficients within a certain range through thresholds. Finally, the original signal is restored through wavelet reconstruction, and the characteristic parameter set of the audible sound signal is obtained.

[0032] (4) For the arc voltage signal, in order to avoid the occurrence of modal aliasing, the welding electrical signal is decomposed by using the adaptive noise complete set empirical mode decomposition to obtain a series of intrinsic mode functions (IMFs). Based on the Pearson correlation coefficient method, the correlation between each IMF and welding defects is quantitatively analyzed, and the IMF components closely related to welding defects are determined. The characteristic parameters of the IMF components are extracted, and the characteristic parameter set of the arc voltage signal is constructed.

[0033] (5) For vibration signals, vibration sensors are used to collect simulated acceleration signals during the arc material addition and subtraction process. The Diagnostic Feature Designer tool in MATLAB is used to extract statistical features from the vibration data, namely RMS, kurtosis, and crest factor, to form a set of characteristic parameters of the vibration signal.

[0034] S3, constructing an SVM classification model, that is, obtaining a comprehensive feature vector by performing data fusion on feature parameters;

[0035] In this step, according to the SVM (support vector machine) classification model established based on the data-driven hybrid Filter and Wrapper, the Filter filter is used to sort the extracted feature parameters; through the Wrapper wrapper, the average classification accuracy is calculated (that is, the mean classification accuracy is obtained), and the optimal feature parameter subset is determined according to the average classification accuracy of the forming defects to form a comprehensive feature vector, thereby realizing the construction of the SVM classification model.

[0036] S3.1. Calculate the Fisher distance of the extracted feature parameters using the Filter filter and sort them according to the Fisher distance;

[0037] Fisher distance criterion is one of the most widely used criteria in feature selection due to its good performance. Its purpose is to separate data of different categories and aggregate data of the same category. This helps us distinguish more clearly and accurately in classification tasks. According to the overall sensitivity of each feature parameter to the product forming defect, each feature parameter is sorted by using Fisher distance criterion.

[0038] definition is the sequence number of the defect type, , is the total number of defect types; Class defects, including training samples, each of which contains characteristic parameters, The characteristic parameter set of the training samples of class defects is , For the Type of defect The first training sample feature parameters.

[0039] No. Prior probability of defect type The calculation formula is:

[0040] ;

[0041] No. Class defect type characteristic parameters The class mean The calculation formula is:

[0042] ;

[0043] Therefore, for all types of defects, the characteristic parameters The mean The calculation formula is:

[0044] ;

[0045] No. Characteristic parameters of defect types The sample covariance matrix of The calculation formula is:

[0046] ;

[0047] In the formula, For the Characteristic parameters of defect types The sample covariance matrix of .

[0048] No. Characteristic parameters of defect types The intra-class scatter matrix and Characteristic parameters of defect types The between-class scatter matrix The calculation formula is as follows:

[0049] ;

[0050] ;

[0051] in, Indicates the compactness of characteristic parameter samples of the same defect type. If the intra-class scatter matrix is ​​small, it means that the sample distribution is relatively concentrated. It reflects the distance between the class means of the characteristic parameter sets of each welding defect type. If the inter-class scatter matrix is ​​larger, it means that the class means between the types are more widely distributed and the separability of the welding defect types is better.

[0052] Therefore, when calculating each characteristic parameter When the Fisher distance (linear discriminant analysis) is used, the Fisher distance calculation formula is:

[0053] ;

[0054] Among them, in order to prevent the relevant important characteristic parameters from being filtered, the weighting factor is set according to the needs of the material addition and subtraction process. , where the weighting factor The value is (0-1).

[0055] According to the calculated Fisher distance value, the feature parameters extracted from each signal are sorted. A higher Fisher value means that the feature parameter can better distinguish the forming defects in the SVM classification model.

[0056] S3.2: For the feature parameters sorted based on the Fisher distance value, the cross-validation calculation method is used to determine the optimal feature subset according to the value of the average classification accuracy of the Wrapper wrapper to form a comprehensive feature vector.

[0057] The classification defect category corresponding to the sample is obtained from the classification decision function value, and the average classification accuracy is calculated based on the actual defects of the sample. According to the average classification accuracy, the feature parameters are selected to form a comprehensive feature vector of the feature parameters to realize the construction of the SVM classification model.

[0058] First, the classification decision function The calculation formula is as follows:

[0059] ;

[0060] in, For the test sample, For the Comparison samples, is the category label of the comparison sample, To compare the sample numbers, is the total number of comparison samples, is a bias term used to adjust the position of the decision boundary to ensure the best classification effect. is the Lagrange multiplier, representing the weight of each comparison sample; is the kernel function, representing the test sample Compare with samples The similarity of , sgn is the sign function, which is a basic function in mathematics and is used to indicate the positive and negative signs of the value of a number; The values ​​obtained correspond to various classification defect categories. According to the expression, The obtained value is 1 or -1, and these two values ​​represent two different types of defects, and the defects are classified accordingly. When more than two types of defects are to be classified, One value of represents a certain defect, and the other value represents the remaining other types of defects. If the specific defect category is not obtained, the remaining defect categories are divided again, and the above method is repeated for multiple classifications until the specific defect type is obtained. The data processed by the defect classification and identification of arc augmentation and subtraction is nonlinearly separable data, so the invention uses the radial basis function (RBF) as the kernel function, then The calculation formula is as follows:

[0061]

[0062] in, is a parameter that controls the width of the Gaussian kernel, is a natural number. The accuracy of the SVM classification model in classifying the data set depends on the parameter The size of (too large a value will lead to overfitting, too small a value will lead to underfitting) is verified by cross-validation. Search for the optimal value of

[0063] according to The value is used to obtain the classification defect category corresponding to the sample, and the average classification accuracy is calculated based on the actual defects of the sample. According to the average classification accuracy, the feature parameters are selected to form a comprehensive feature vector of the feature parameters to realize the construction of the SVM classification model.

[0064] S4. Divide the training set and the verification set according to the feature parameters extracted in real time, input the training set into the SVM classification model for training, and verify the effectiveness of the SVM classification model according to the product forming defects;

[0065] The characteristic parameters are extracted from the processed product signals collected in real time; the characteristic parameter samples are randomly divided into subsamples of equal size. Among the sub-samples, one sub-sample is reserved as the validation data for the test model, and the remaining subsamples as training data. Then repeat the cross validation process Second-rate, Each of the subsamples is used only once as validation data.

[0066] S5. Combining the trained SVM classification model with the defect classification and recognition set is used for arc additive and subtractive forming defect classification and recognition, and obtaining defect classification and recognition results in the arc additive and subtractive forming process;

[0067] S6. According to the defect classification and identification results of the SVM classification model, the defective parts of the product are reduced in real time through the milling system, and the arc additive manufacturing related processing parameters are adjusted in time by using the model-free adaptive iterative learning control algorithm to increase or reduce the material again;

[0068] The trained SVM model is used to monitor the state of the additive and subtractive materials in arc additive and subtractive manufacturing, and to diagnose defects. For defects that have already occurred, the milling system is used to remove the defective parts, and the model-free adaptive iterative learning control algorithm is used to iteratively adjust the processing parameters in real time according to the diagnosed defects. An equivalent linearized data model is established at each working moment, and the pseudo partial derivative is updated online according to the system I / O data. The input of parameters related to product forming such as the droplet height and width in the arc additive and subtractive manufacturing process is adjusted by solving the following equations.

[0069] The input and output system of parameters related to product forming in arc additive and subtractive manufacturing is defined as:

[0070] ;

[0071] The system does not rely on the precise mathematical model of the system, but directly improves the control performance by updating the error based on the input and output data. For the Iterations The output at the time, for Output and input function at each moment; No. Iterations The output at the time, For the Iterations The input amount at time, represents the iteration index, Indicates time.

[0072] Then the change between input and output can be expressed as:

[0073] ;

[0074] ;

[0075] In the formula, For the Iterations The output at the time, For the Second and The output error between iterations, For the Iterations The input amount at time, For the Second and The error in the input between iterations.

[0076] Assume Second and The output error between iterations is Second and The relationship between the input errors between iterations is:

[0077] ;

[0078] Among them, The pseudo-partial derivative of the iteration is a time-varying, iteration-dependent parameter named pseudo partial derivative. is unknown and needs to be estimated. We use The estimated value of the pseudo-partial derivative of the iteration To represent the pseudo partial derivative . It can be obtained by the following equation, that is, The pseudo-partial derivative estimate of the iteration Can be obtained by The estimated value of the iteration Find:

[0079] ;

[0080] in, For the The pseudo partial derivative of the iteration, For the Second and The input error between iterations, For the Second and The output error between iterations.

[0081] Initial values ​​for pseudopartial derivative estimates It can be determined based on experience. In the absence of any prior knowledge of the system, the initial value of the pseudo partial derivative can be simply set to 0. If there is some prior knowledge of the system, an initial value based on experience can be used.

[0082] No. Second and The input error between iterations The calculation formula is:

[0083] ;

[0084] in, ;

[0085] In the formula, The value of , is a positive weight coefficient, For the The control input for the iteration is is the step factor, is the weight factor, is the step size factor, is the expected output, It is The system output of the iteration is is the tracking error, i.e. The difference between the actual output and the expected output of the iteration.

[0086] In terms of the entire model, the control input of the next iteration is optimized by introducing errors in the previous iteration, where the input is the parameters related to product forming, such as the droplet height and width.

[0087] The invention discloses an arc additive and subtractive forming defect diagnosis and intelligent control system based on data fusion, which comprises an infrared thermal imager, a Hall sensor, an acoustic sensor, a spectrometer, a vibration sensor, a computer and a milling system.

[0088] Multimodal information in arc additive and subtractive manufacturing is acquired through infrared thermal imagers, Hall sensors, acoustic sensors, spectrometers, and vibration sensors, and the acquired information is transmitted to a computer, which performs relevant data processing to diagnose forming defects in additive and subtractive manufacturing. The resulting product defects are removed by a cutting and milling system, and the processing parameters are adjusted in time through a model-free adaptive iterative learning control algorithm, and the material is added and subtracted again to obtain processed parts with good forming quality.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the essence and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for arc-assisted material forming defect diagnosis and intelligent control based on data fusion, characterized in that: S1. Use infrared thermal imagers, Hall sensors, acoustic sensors, spectrometers, and vibration sensors to obtain multi-modal data signals in arc additive and subtractive manufacturing in real time; S2. Extract characteristic parameters in real time from the acquired molten pool signal, audible sound signal, arc voltage signal, vibration signal and arc spectrum signal; wherein, for the molten pool signal, a computational heat transfer method based on graph theory is used to predict the temperature distribution of the component, and the average value and standard deviation of the temperature predicted by the graph theory method will be used as characteristic parameters of the molten pool signal; S3, constructing an SVM classification model, that is, obtaining a comprehensive feature vector by performing data fusion on feature parameters; S4. Divide the training set and the verification set according to the feature parameters extracted in real time, input the training set into the SVM classification model for training, and verify the effectiveness of the SVM classification model according to the product forming defects; S5. Combining the trained SVM classification model with the defect classification and recognition set is used for arc additive and subtractive forming defect classification and recognition, and obtaining defect classification and recognition results in the arc additive and subtractive forming process; S6. According to the defect classification and identification results of the SVM classification model, the defective parts of the product are reduced in real time through the cutting and milling system, and the relevant processing parameters of arc additive and subtractive manufacturing are adjusted in time using the model-free adaptive iterative learning control algorithm to re-add and subtract materials.

2. The arc-assisted material forming defect diagnosis and intelligent control method according to claim 1 is characterized in that: In S2, a computational heat transfer method based on graph theory is used to predict the temperature distribution of the component for the molten pool signal. The average value and standard deviation of the temperature predicted by the graph theory method will be used as the characteristic parameters of the molten pool signal. The temperature thermal characteristics predicted by the graph theory are combined with the thermal characteristics obtained by the thermal imager in the arc additive manufacturing process for defect diagnosis. Graph theory thermal modeling is to solve the continuous heat diffusion equation. The calculation formula for the spatiotemporal temperature distribution L(q,z,v,t) at each position (q,z,v) inside the part at time t is: Where t is time, τ is density, C p is the specific heat capacity, o is the thermal conductivity, (q,z,v) is the three-dimensional coordinate value inside the part, and the right side of the equation is the heat input Q at the time position (q,z,v); in arc additive manufacturing, the heat input Q is the energy provided by the arc.

3. The arc-assisted material forming defect diagnosis and intelligent control method according to claim 1, characterized in that: The S3 step specifically includes: S3.1, using the filter to calculate the Fisher distance of the extracted feature parameters, and sorting them according to the Fisher distance; S3.

2. For the feature parameters sorted based on the Fisher distance value, the optimal feature subset is determined according to the value of the average classification accuracy of the encapsulator using the cross-validation calculation method to form a comprehensive feature vector.

4. The arc-assisted material forming defect diagnosis and intelligent control method according to claim 3 is characterized in that: In S3.1, the calculation formula of Fisher distance for each feature parameter l is: Among them, θ is the weighting factor, S b is the inter-class scatter matrix of the characteristic parameter l of the i-th defect type, S′ b is the intra-class scatter matrix of the characteristic parameter l of the i-th defect type; i is the serial number of the defect type, and l is the characteristic parameter contained in the training sample.

5. The arc-assisted material forming defect diagnosis and intelligent control method according to claim 4, characterized in that: The inter-class scatter matrix S of the characteristic parameter l of the i-th defect type b The calculation formula is: Where i is the serial number of the defect type, i = (1, 2, ..., C), C is the total number of defect types, is the class mean of the characteristic parameter l of the i-th defect type, is the mean value of the characteristic parameter l of all types of defects; The intra-class scatter matrix S′ of the characteristic parameter l of the i-th defect type b The calculation formula is: Where P i is the prior probability of the i-th defect type; is the sample covariance matrix of the characteristic parameter l of the i-th defect type.

6. The arc-assisted material forming defect diagnosis and intelligent control method according to claim 4, characterized in that: In S3.2, the classification decision function value is used to obtain the classification defect category corresponding to the sample, and the average classification accuracy is calculated in combination with the actual defects of the sample; the feature parameters are selected according to the average classification accuracy, and then a comprehensive feature vector of the feature parameters is formed to realize the construction of the SVM classification model: the classification decision function f(x) is calculated as follows: In the formula, x is the test sample, x w is the wth comparison sample, y w is the category label of the comparison sample, w is the comparison sample serial number, r is the total number of comparison samples, is the bias term, β w is the Lagrange multiplier, representing the weight of each comparison sample; d(x,x w ) is the kernel function, which represents the difference between the test sample x and the comparison sample x w sgn is the sign function, which is a basic function in mathematics and is used to indicate the positive and negative signs of numbers.

7. The arc-assisted material forming defect diagnosis and intelligent control method according to claim 1, characterized in that: In S3.2, the trained SVM model is used to monitor the state of additive and subtractive materials in arc additive and subtractive manufacturing, and to diagnose defects. For defects that have already occurred, the milling system is used to remove the defective parts, and the model-free adaptive iterative learning control algorithm is used to iteratively adjust the processing parameters in real time according to the diagnosed defects; an equivalent linearized data model is established at each working moment, and the pseudo-partial derivative is updated online according to the system I / O data. The input quantities of parameters related to product forming, such as the droplet height and width, in the arc additive and subtractive manufacturing process are adjusted by solving the following equations.

8. The arc-assisted material forming defect diagnosis and intelligent control method according to claim 7, characterized in that: The input and output system of parameters related to product forming in arc additive and subtractive manufacturing is: where y k (t+1) is the output at the kth iteration (t+1), is the function of output and input at time t; y k (t) The output at time t of the kth iteration, u k (t) is the input at time t of the kth iteration, k represents the iteration index, and t represents the time; Then the change between input and output is expressed as: Δy k (t+1)=y k (t+1)-y k-1 (t+1); Δu k (t)=u k (t)-u k-1 (t); In the formula, y k-1 (t+1) is the output at the k-1th iteration (t+1), Δy k (t+1) is the output error between the kth and k-1th iterations, u k-1 (t) is the input at time t of the k-1th iteration, Δu k (t) is the input error between the kth and k-1th iterations.

9. The arc-assisted material forming defect diagnosis and intelligent control method according to claim 8, characterized in that: The relationship between the output error between the kth and k-1th iterations and the input error between the kth and k-1th iterations is defined as: Δy k (t+1)=φ k (t)D k (t); where φ k (t) is the pseudo partial derivative of the kth iteration, and the estimated value of the pseudo partial derivative of the kth iteration is Conduct valuations; The estimated value of the pseudo-partial derivative at the kth iteration The calculation formula is: in is the estimated value of the pseudo partial derivative of the k-1th iteration, Δu k-1 (t) is the input error between the k-1th and k-2th iterations, Δy k-1 (t+1) is the output error between the k-1th and k-2th iterations; Δu of the input error between the kth and k-1th iterations k (t) is calculated as: In the formula, φ k The value of (t) is given by The value is replaced by μ, μ is the positive weight coefficient, u k (t) is the control input, ρ is the step factor, λ is the weight factor, η is the step size factor, y h (t+1) is the expected output, y k (t+1) is the system output of the kth iteration, e k-1 (t+1) is the tracking error, which is the difference between the actual output and the expected output at the k-1th iteration.

10. An arc-assisted material forming defect diagnosis and intelligent control system based on data fusion, characterized in that: The method for diagnosing and intelligently controlling arc additive and subtractive material forming defects according to any one of claims 1 to 9, wherein the system comprises an infrared thermal imager, a Hall sensor, an acoustic sensor, a spectrometer, a vibration sensor, a computer, and a milling system; multimodal information in arc additive and subtractive material manufacturing is obtained by means of the infrared thermal imager, the Hall sensor, the acoustic sensor, the spectrometer, and the vibration sensor, and the collected information is transmitted to the computer, which performs relevant data processing to diagnose forming defects in additive and subtractive material manufacturing, remove the generated product defects by means of the milling system, and adjust the processing parameters in time by means of a model-free adaptive iterative learning control algorithm, and re-add and subtract materials to obtain formed processed parts.

Images