Welding machine temperature upper and lower limit reduction method based on welding historical data

Abstract

The invention discloses a welding historical data-based welding machine temperature upper and lower limit reduction method, which comprises the following steps of: realizing C equivalent calculation according to chemical components on the basis of historical welding actual performance data, and establishing a deep neural network model by taking the C equivalent, the thickness of to-be-welded strip steel and key welding parameters as input and the actual temperature as output; the temperature deviation is given according to the evaluation result, so that the upper limit and the lower limit of the temperature are determined. According to the thickness of to-be-welded strip steel, C equivalent and welding process parameters, a deep neural network model is established, and upper and lower limits of temperature are optimized.

Description

Technical Field

[0001] The present invention relates to a thin strip steel welding control technology, and more particularly to a method for reducing upper and lower limits of welding machine temperature based on welding history data. Background Art

[0002] Currently, the upper and lower temperature limits of cold-rolled strip lap welding machines are concentrated within ±125°C. This wide temperature range prevents actual temperature fluctuations from exceeding this limit. This range can only be exceeded if welding parameters, especially the welding current, are incorrectly set. This wide range makes temperature fluctuations insensitive, leading to the risk of missed detections.

[0003] When a new steel grade is introduced, it's necessary to conduct real-time simulation experiments on the welding process parameters for the cold-rolled strip being welded, either during a mill shutdown or directly during the mill shutdown, supplemented by manual inspection. In this case, optimizing the upper and lower welding temperature limits not only provides data support for weld quality assessments but also contributes to optimizing welding parameters.

[0004] Furthermore, during welding, the carbon equivalent of the materials being welded has a direct impact on weld quality. However, the impact of carbon equivalent has been rarely considered in lap welding of cold-rolled strip. Therefore, by using carbon equivalent as one of the input parameters, the impact of carbon equivalent on weld quality can be incorporated into the model. Summary of the Invention

[0005] In response to the defects existing in the prior art, the purpose of the present invention is to provide a method for reducing the upper and lower limits of welding machine temperature based on welding history data. According to the thickness of the steel strip to be welded, the C equivalent, and the welding process parameters, a deep neural network model is established to optimize the upper and lower limits of temperature.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for reducing the upper and lower limits of welding machine temperature based on welding history data:

[0008] Based on historical welding performance data, the C equivalent is calculated according to the chemical composition. Then, the C equivalent, the thickness of the steel strip to be welded, and key welding parameters are used as input, and the actual temperature is used as output. A deep neural network model is established to evaluate the temperature. Based on the evaluation results, a temperature deviation is given to determine the upper and lower temperature limits.

[0009] Preferably, the method for reducing the upper and lower limits of the welding machine temperature specifically includes the following steps:

[0010] S1, collect historical welding performance data;

[0011] S2, based on historical welding performance data, obtains the chemical composition of the material, including C, Mn, Cr, Cu, Mo, V, and Ni;

[0012] S3, calculating the C equivalent of the two strips of the current weld according to the C equivalent calculation formula and chemical composition;

[0013] S4, establish a deep neural network model, taking the C equivalent, thickness and welding performance data of the two strips as input and the actual temperature as output;

[0014] S5, dividing the training set and the prediction set, using the training set data as the training of the deep neural network model, and using the prediction set as the result verification of the deep neural network model training, that is, outputting the predicted temperature;

[0015] S6, based on the difference distribution between the predicted temperature and the actual temperature as a judgment basis, establish the deviation band of the upper and lower temperature limits, and based on the on-site production experience, use this deviation band as the optimized range of the upper and lower temperature limits;

[0016] S7. After welding, the upper and lower temperature limits are determined based on the tapping marks and thickness of the two strips of the weld, and compared with the actual welding temperature to determine the welding temperature.

[0017] Preferably, the historical welding performance data includes the steel tapping mark, thickness, material chemical composition, and corresponding welding process parameters and welding temperature of the preceding and succeeding coils.

[0018] Preferably, the welding process parameters include welding current, welding speed, welding wheel pressure, overlap amount, and overlap compensation amount.

[0019] Preferably, in step S3, the C equivalent calculation formula is:

[0020] CE=C+Mn / 6+(Cr+Mo+V) / 5+(Ni+Cu) / 15(%).

[0021] Preferably, in step S6, the difference distribution between the predicted temperature and the actual temperature is used as a basis for determination, and deviation bands of the upper and lower temperature limits are established and applied to the optimization of the upper and lower temperature limits.

[0022] The present invention provides a method for reducing the upper and lower limits of welding machine temperature based on welding history data, which has the following beneficial effects:

[0023] 1) Establish a deep neural network model to optimize the upper and lower temperature limits, narrow the temperature range, provide new criteria for weld quality judgment, and reduce the possibility of missed detection. At the same time, optimize the upper and lower limit ranges, making it more sensitive to welding parameters and able to detect any deviations in welding parameters;

[0024] 2) Consider the effect of C equivalent on welding results and incorporate it into the model along with welding parameters as input. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 Schematic diagram of the principle of the method for reducing the upper and lower limits of the welding machine temperature according to the present invention;

[0026] Figure 2 It is a histogram of the difference between the predicted temperature and the actual temperature of the weld in the method for reducing the upper and lower limits of the welding machine temperature of the present invention;

[0027] Figure 3 It is a flow chart of the DNN neural network algorithm in the method for reducing the upper and lower limits of welding machine temperature of the present invention. DETAILED DESCRIPTION

[0028] In order to better understand the above technical solutions of the present invention, the technical solutions of the present invention are further described below with reference to the accompanying drawings and embodiments.

[0029] The present invention provides a method for reducing the upper and lower limits of welding machine temperature based on welding history data:

[0030] Based on historical welding performance data, the C equivalent is calculated according to the chemical composition. Then, the C equivalent, the thickness of the steel strip to be welded, and key welding parameters are used as input, and the actual temperature is used as output. A deep neural network model is established to evaluate the temperature. Based on the evaluation results, a temperature deviation is given to determine the upper and lower temperature limits.

[0031] The method for reducing the upper and lower limits of the welding machine temperature specifically includes the following steps:

[0032] S1, collecting historical welding performance data, including the steel tapping mark, thickness, material chemical composition of the previous and next coils, and the corresponding welding process parameters and welding results (temperature). The welding process parameters include welding current, welding speed, welding wheel pressure, overlap, and overlap compensation;

[0033] S2, based on historical welding performance data, obtains the chemical composition of the material, including C, Mn, Cr, Cu, Mo, V, and Ni;

[0034] S3, calculating the C equivalent of the two strips of the current weld according to the C equivalent calculation formula and chemical composition;

[0035] S4, establish a deep neural network model, take the C equivalent, thickness and welding performance data (welding current, welding speed, welding wheel pressure, overlap, overlap compensation) of the two strips as input, and use the welding result (temperature) as output;

[0036] S5, dividing the training set and the prediction set, using the training set data as the training of the deep neural network model, and using the prediction set as the result verification of the deep neural network model training, that is, outputting the predicted temperature;

[0037] S6, based on the difference distribution between the predicted temperature and the actual temperature as a judgment basis, establish the deviation band of the upper and lower temperature limits, and based on the on-site production experience, use this deviation band as the optimized range of the upper and lower temperature limits;

[0038] S7, after welding, the temperature upper and lower limits are corresponding to the tapping marks and thickness of the two strips of the weld, and compared with the actual welding temperature to determine the welding result (temperature).

[0039] In the above step S3, the C equivalent calculation formula is:

[0040] CE=C+Mn / 6+(Cr+Mo+V) / 5+(Ni+Cu) / 15(%).

[0041] In the above step S6, the difference distribution between the predicted temperature and the actual temperature is used as a basis for determination, and the deviation bands of the upper and lower temperature limits are established and applied to the optimization of the upper and lower temperature limits.

[0042] The system used in the method for reducing the upper and lower limits of the welding machine temperature of the present invention is as follows Figure 1 As shown, it includes welding machine system, supporting software system (OLW), and corresponding data services and support (such as digital steel coil data platform).

[0043] Here's how it works:

[0044] Using the on-site production data collected by the digital steel coil as the carrier, the steel tapping marks, thickness, chemical composition (C, Mn, Cr, Cu, Ni, Mo, V, etc.) of the strip at the uncoiler (the rear coil strip) and the running strip (the front coil strip) are recorded, and then the C equivalent is calculated according to the C equivalent calculation formula; with thickness, C equivalent, and welding process parameters as input and temperature as output, a deep neural network model is established to optimize the upper and lower limits of temperature.

[0045] In the existing technology framework, the upper and lower temperature limits have large deviations, are insensitive to parameter changes and temperature fluctuations, and may miss detections. This invention processes historical data and establishes a deep neural network model to reduce and optimize the upper and lower temperature limits.

[0046] Obtain the chemical composition of the front and rear coils, and calculate the carbon equivalent based on the carbon equivalent calculation formula, taking the chemical composition as input. The chemical composition of the front and rear coils includes C, Mn, Cr, Cu, Mo, V, and Ni; the calculation formula for carbon equivalent is:

[0047] CE=C+Mn / 6+(Cr+Mo+V) / 5+(Ni+Cu) / 15(%)

[0048] Based on the difference distribution between the predicted temperature and the actual temperature, a deviation band of the upper and lower temperature limits is established, and based on on-site production experience, this deviation band is used as the optimized range of the upper and lower temperature limits. After welding, the temperature and upper and lower temperature limits are corresponding to the steel tapping marks and thickness of the two strips of the weld, and compared with the actual welding temperature to determine the welding result (temperature).

[0049] Feedforward neural networks are widely used and mature in current industrial fields. The BP neural network is a typical feedforward neural network. In 1986, scientists such as Rumelhart, Hinton, and Williams first proposed the concept of the BP neural network model. Today, BP neural networks are mature and widely used. Due to their simple structure, easy operation, and strong nonlinear mapping capabilities, they have become one of the most widely used neural networks in engineering.

[0050] The BP neural network is a feedforward neural network in which the error propagates backward from the output layer. The basic principle is to use sample data, transforming it layer by layer from the input layer through the hidden layers, ultimately obtaining output data at the output layer. The resulting output data is compared with the true value, and the error between the two is distributed to each node in each layer to modify the weights and thresholds of each node. This forward and backward propagation process is repeated, ultimately transforming the network into a mapping network model that achieves the desired results. The BP neural network consists of an input layer, a hidden layer, and an output layer, each composed of multiple neurons. The number of nodes in the input layer matches the dimension of the input vector; the number of nodes in the output layer matches the dimension of the preset label vector. There is no specific requirement for the number of nodes in the hidden layer; it is usually set using empirical formulas. The hidden layer can be single or multi-layer, but generally does not exceed two. The input, hidden, and output layers are interconnected. Nodes within each layer are not interconnected, but nodes between layers are fully connected.

[0051] By setting the network depth and number of nodes in a BP neural network, a deep neural network (DNN) model can be generated. The DNN learning process consists of two phases: forward propagation of the input signal and back propagation of the error. Input data is transmitted to the output through forward propagation, where it is then compared with the actual data. If the output data does not match the expected data, the error is distributed across each node in each layer. The network is then updated by modifying the thresholds of each node and the weights between nodes in each layer. DNNs gradually approach the desired model through this continuous iterative update process.

[0052] The specific algorithm steps of the DNN neural network are:

[0053] (1) Determine the DNN neural network structure and network parameter settings. Before network training, the designed network structure needs to be constructed, and appropriate network parameters must be set to ensure that the training network can meet the expected requirements.

[0054] (2) Initialization of weights and thresholds: DNN networks that do not use a specific algorithm to set weights and thresholds are randomly assigned a set of small non-zero values.

[0055] (3) Process the training sample data and send it to the network for training. Before sending it to the network for training, the training sample data of different orders of magnitude need to be normalized and then sent to the network to start training.

[0056] (4) Forward propagation stage: Calculate the output of each layer in sequence and obtain the final output Ym at the output layer.

[0057] (5) Compare the actual output with the expected output. If the actual output Ym is consistent with the expected output Om, the training ends; if the actual output Ym is inconsistent with the expected output Om, the back propagation phase begins.

[0058] (6) Back propagation stage. Calculate the total error E between the actual output Ym and the expected output Om, distribute the total error E to each node in each layer, and adjust the weight and threshold of each node in each layer.

[0059] (7) The updated network undergoes the next training. The network with adjusted weights and thresholds is used as a new model for training again. If the training result meets the expected requirements, the training ends. If not, the error back propagation phase is entered again. The cycle repeats until the number of iterations reaches the preset number or the performance function value is less than the preset error accuracy, then the training process stops. The algorithm flow of the DNN neural network is as follows: Figure 3 shown.

[0060] The 0LW software communicates with the welding machine and sends the judgment results to the welding machine PLC after welding. The welding machine PLC obtains the parameters, displays them on the welding machine operation terminal, and performs the corresponding welding.

[0061] Example

[0062] This example obtains the chemical composition of the front and rear steel coils and calculates the C equivalent based on the C equivalent calculation formula, taking the chemical composition as input. The chemical composition of the front and rear steel coils includes C, Mn, Cr, Cu, Mo, V, and Ni; the calculation formula for C equivalent is:

[0063] CE=C+Mn / 6+(Cr+Mo+V) / 5+(Ni+Cu) / 15(%)

[0064] The calculation results of C equivalent are shown in Table 1 below.

[0065] Table 1 Calculation results of some tapping symbols C equivalent

[0066] The first six digits of the tapping mark First six C equivalent AN0543 120543 0.072272919 AN0691 120691 0.081770971 AN0891 120891 0.114100772 AN1053 121053 0.101605655 AN1064 121064 0.107455852 AN1111 121111 0.078671776 AN1122 121122 0.078976809 AN1540 121540 0.096343528 AN1950 121950 0.140293371 AN2222 122222 0.054047103 AN2233 122233 0.070794031 AN4444 124444 0.106337912 AN6666 126666 0.028612841 AN8888 128888 0.056978571 AP0540 130540 0.070372042 AP0941 130941 0.083058716

[0067] As shown in Table 2 below, historical welding performance data was collected, including the tapping marks of the preceding and succeeding coils, thickness welding process parameters, and temperature.

[0068] Table 2 Welding performance data

[0069]

[0070] The C equivalent in Table 1 and the thickness and welding process parameters in Table 2 are integrated as input, and the temperature is used as the output to establish a three-layer (the number of nodes in each layer is 9:9:15) deep neural network model. In the training process of the DNN model, the 'ogsig', 'tansig', 'purel in' activation functions and the 'trainlm' training function are used. The error threshold, initial learning rate and number of iterations are set to 0.000001, 0.01 and 1000 respectively. The data of the first six months (excluding NG welds) are used as the total data set. The first five months are used as the training set, and the last month is used as the test set. The total amount of test set data is 2385. The input of the model is the thickness of the front and rear rolls, C equivalent, and the average of welding parameters (current, speed, etc.), and the output is the average temperature. The difference between the predicted temperature and the actual temperature average is used as the horizontal axis, and a bar chart is drawn as shown below. Figure 2 The analysis results are shown in Table 3 below.

[0071] Table 3 Temperature deviation table

[0072] Deviation 1 Quantity 1 Proportion 1 -50＜＝t＜＝50 2319 97.23％ the remaining 66 2.77％ Deviation 2 Quantity 2 2% -60＜＝t＜＝60 2360 98.95％ the remaining 25 1.05％

[0073] As shown in Table 3, the temperature upper and lower limit deviations are within the range of -60<=t<=60, accounting for 98.95%, so the temperature upper and lower limit range is set within -60<=t<=60°C.

[0074] Then, based on the first six digits of the tapping mark and the thickness, the welds were classified, with one tapping mark and one thickness range (as shown in Table 4). The mean and standard deviation of the temperature within this category were calculated, and data within the range of ±3 standard deviations of the mean were selected to eliminate abnormal data. The mean of the data after eliminating abnormal data was then taken as the true temperature value for this category, and the upper limit of the temperature was set to +60°C, and the lower limit was set to -60°C (as shown in Table 5).

[0075] Table 4 Thickness corresponding interval

[0076] Thickness classification Set thickness 0.3 0.27＜t≤0.33 0.4 0.36＜t≤0.44 0.5 0.45＜t≤0.55 0.6 0.55＜t≤0.65 0.7 0.65＜t≤0.75 0.8 0.75＜t≤0.90 1.0 0.90＜t≤1.1 1.2 1.1＜t≤1.3 1.4 1.3＜t≤1.5 1.6 1.5＜t≤1.7 1.8 1.7＜t≤1.9 2.0 1.9＜t≤2.2 2.3 2.2＜t

[0077] Table 5 Actual values ​​and upper and lower limits of temperature for some categories

[0078]

[0079] Those skilled in the art should recognize that the above embodiments are merely intended to illustrate the present invention and are not intended to limit the present invention. As long as they are within the spirit of the present invention, any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention.

Claims

1. A method for reducing the upper and lower limits of welding machine temperature based on welding history data, characterized in that: Based on historical welding performance data, the C equivalent is calculated according to the chemical composition. Then, the C equivalent, the thickness of the steel strip to be welded, and key welding parameters are used as input, and the actual temperature is used as output. A deep neural network model is established to evaluate the temperature. Based on the evaluation results, a temperature deviation is given to determine the upper and lower temperature limits.

2. The method for reducing the upper and lower limits of welding machine temperature based on welding history data according to claim 1, characterized in that: The method for reducing the upper and lower limits of the welding machine temperature specifically comprises the following steps: S1, collect historical welding performance data; S2, based on historical welding performance data, obtains the chemical composition of the material, including C, Mn, Cr, Cu, Mo, V, and Ni; S3, calculating the C equivalent of the two strips of the current weld according to the C equivalent calculation formula and chemical composition; S4, establish a deep neural network model, taking the C equivalent, thickness and welding performance data of the two strips as input and the actual temperature as output; S5, dividing the training set and the prediction set, using the training set data as the training of the deep neural network model, and using the prediction set as the result verification of the deep neural network model training, that is, outputting the predicted temperature; S6, based on the difference distribution between the predicted temperature and the actual temperature as a judgment basis, establish the deviation band of the upper and lower temperature limits, and based on the on-site production experience, use this deviation band as the optimized range of the upper and lower temperature limits; S7. After welding, the upper and lower temperature limits are determined based on the tapping marks and thickness of the two strips of the weld, and compared with the actual welding temperature to determine the welding temperature.

3. The method for reducing the upper and lower limits of welding machine temperature based on welding history data according to claim 2, characterized in that: The historical welding performance data includes the steel tapping mark, thickness, material chemical composition, and corresponding welding process parameters and welding temperature of the preceding and succeeding coils.

4. The method for reducing the upper and lower limits of welding machine temperature based on welding history data according to claim 3, characterized in that: The welding process parameters include welding current, welding speed, welding wheel pressure, overlap amount, and overlap compensation amount.

5. The method for reducing the upper and lower limits of welding machine temperature based on welding history data according to claim 2, characterized in that: In step S3, the C equivalent calculation formula is: CE=C+Mn / 6+(Cr+Mo+V) / 5+(Ni+Cu) / 15(%).

6. The method for reducing the upper and lower limits of welding machine temperature based on welding history data according to claim 2, characterized in that: In step S6, the difference distribution between the predicted temperature and the actual temperature is used as a basis for determination, and deviation bands of the upper and lower temperature limits are established and applied to the optimization of the upper and lower temperature limits.

Images