Optimization Method for Turning Process of Mechanical Parts in Batch Production Based on Pattern Recognition

Abstract

The present invention proposes an optimization method for the turning process of mechanical parts based on batch production of pattern recognition. The optimization method includes: S100, collecting turning process parameters and determining the mode of the turning process through a pre-trained recognition model; S200, comparing the processing parameters of different workpieces in the same mode to obtain the first optimization criterion, and comparing the processing parameters of the same workpiece in different modes to obtain the second optimization criterion; S300, optimizing the turning process based on the first optimization criterion and the second optimization criterion. The present invention divides the processing process into different stages and uses the trained model to identify different stages. By comparing the processing parameters of different workpieces in the same processing stage and the processing parameters between different processing stages of the same workpiece, once an abnormal / potential optimization point is detected, the system issues a prompt and gives optimization suggestions. Thus, the characteristics of different processing processes can be taken into account, and the turning process can be optimized to improve the turning efficiency.

Description

Technical Field

[0001] The invention relates to the technical field of parts processing, and in particular to a method for optimizing the machining process of mechanical parts produced in batches based on pattern recognition. Background Art

[0002] At present, there are several methods for optimizing the process of machining mechanical parts in mass production:

[0003] 1. Adjust process parameters through experience to improve efficiency;

[0004] 2. Automatically adjust according to the real-time processing load through adaptive algorithms;

[0005] 3. By loading power or vibration sensors, we can manually determine the parts of the processing that can be optimized, and improve efficiency by adjusting the feed.

[0006] The first method mentioned above is highly dependent on experience and is not a systematic method to improve the processing process.

[0007] The second method is that some manufacturers on the market currently provide smart boxes loaded with adaptive algorithms. Through the system's real-time detection of load changes, the feed rate is adjusted adaptively. However, since it is based on a general algorithm and is not subdivided for specific processing scenarios, the efficiency improvement is generally not high, at 5%-10%. For mass-produced mechanical parts, the efficiency improvement is not significant.

[0008] The third method is very demanding for the implementer. It requires not only that the implementer has a good understanding of the machining process of mechanical parts for mass production, but also that the implementer has strong data processing capabilities. In addition, although the selection and judgment of the optimization point is based on the fluctuation of power or vibration during the processing (for example, if there is a period of relatively stable vibration, the feed speed of this period of processing will be increased), this method still relies heavily on experience, and the discovery and optimization of improvement points are not systematic. Summary of the invention

[0009] The technical problem to be solved by the present invention is how to reasonably optimize the lathe machining process. The present invention proposes a method for optimizing the lathe machining process of mechanical parts in batch production based on pattern recognition.

[0010] A method for optimizing a lathe process based on pattern recognition according to an embodiment of the present invention includes:

[0011] S100, collecting machining parameters and determining the machining mode through a pre-trained recognition model;

[0012] S200. Obtain the first optimization criterion by comparing the processing parameters of different workpieces in the same mode, and obtain the second optimization criterion by comparing the processing parameters of the same workpiece in different modes;

[0013] S300. Optimize the turning process based on the first optimization criterion and the second optimization criterion.

[0014] According to the turning process optimization method based on pattern recognition in the embodiments of the present invention, by studying the specific processing process, the present invention divides the processing process into different stages and uses the trained model to identify different stages. By comparing the same processing stage and different processing stages, once an abnormal / potential optimization point is detected, the system will give a prompt and an optimization suggestion. Thus, the characteristics of different processing processes can be taken into account, and the turning process can be systematically optimized, which can better improve the turning efficiency.

[0015] According to some embodiments of the present invention, the method further includes:

[0016] Pre-divide the turning process into multiple modes, and train through a neural network model with the preset parameters in the turning process to obtain the recognition model for identifying each mode.

[0017] In some embodiments of the present invention, the neural network model uses LSTM, and the preset parameters include: power, feed speed in the X-axis direction, feed speed in the Z-axis direction, vibration in the X-axis direction of the spindle, vibration in the Z-axis direction of the spindle, vibration in the X-axis direction of the turret, and vibration in the Z-axis direction.

[0018] According to some embodiments of the present invention, in step S200, obtaining the first optimization criterion by comparing the processing parameters of different workpieces in the same mode includes:

[0019] S211. Statistically calculate the processing duration of a preset number of different workpieces in the same mode and determine the duration threshold;

[0020] S212. Adjust and optimize the feed speed in the corresponding mode based on the duration threshold.

[0021] In some embodiments of the present invention, in step S200, obtaining the second optimization criterion by comparing the processing parameters in different modes includes:

[0022] Compare the vibration parameters of the same workpiece in different modes. If the difference in the effective vibration values of the same workpiece in different modes exceeds the first threshold, adjust and optimize the feed speed in the corresponding mode.

[0023] According to some embodiments of the present invention, adjusting and optimizing the feed speed in the corresponding mode includes:

[0024] Increase the feed rate in the corresponding mode with a lower effective vibration value; and / or decrease the feed rate in the corresponding mode higher than the second threshold.

[0025] According to the machining optimization method for mass-produced mechanical parts of an embodiment of the present invention, the machining optimization method uses the above-mentioned turning process optimization method based on pattern recognition to optimize the machining of mass-produced mechanical parts.

[0026] According to the machining optimization method for mass-produced mechanical parts of an embodiment of the present invention, it can take into account the characteristics of different machining processes and systematically optimize turning, and can better improve the turning efficiency of mass-produced mechanical parts.

[0027] According to some embodiments of the present invention, the modes divided in the turning process of mechanical parts include: rapid feed, rough turning of the end face, finish turning of the end face, rapid traverse, rough turning of the outer circle and drilling, rough turning of the remaining outer circle, finish turning of the outer circle, deceleration, and loading and unloading.

[0028] In some embodiments of the present invention, turning parameters are obtained through a newly installed power sensor, vibration sensor, and the PLC of the processing equipment.

[0029] According to some embodiments of the present invention, the machining method is based on the real-time collected turning parameters to perform real-time optimization on the turning process of mechanical parts. Description of the Drawings

[0030] Figure 1 It is a flowchart of the turning process optimization method based on pattern recognition according to an embodiment of the present invention;

[0031] Figure 2 It is a structure diagram of the LSTM neural network according to an embodiment of the present invention;

[0032] Figure 3 It is a training flowchart of the LSTM model according to an embodiment of the present invention;

[0033] Figure 4 It is a flowchart of the method for machining a cross shaft using the turning process optimization method based on pattern recognition according to an embodiment of the present invention;

[0034] Figure 5 It is a schematic diagram of different modes of heavy cross shaft turning according to an embodiment of the present invention;

[0035] Figure 6 It is a schematic diagram of the pattern recognition result according to an embodiment of the present invention. Detailed Embodiments

[0036] To further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined purpose, the present invention will be described in detail as follows in conjunction with the accompanying drawings and preferred embodiments.

[0037] In the present invention, the description of the method flow in the specification and the steps in the flowchart in the accompanying drawings of the present invention do not necessarily need to be strictly executed according to the step numbers. The method steps can change the execution order. Moreover, certain steps can be omitted, multiple steps can be combined into one step for execution, and / or one step can be decomposed into multiple steps for execution.

[0038] As Figure 1 shown, the method for optimizing the turning process based on pattern recognition according to an embodiment of the present invention includes:

[0039] S100, collecting turning parameters and judging the mode in which the turning is located through a pre-trained recognition model;

[0040] For example, the turning process can be pre-divided into multiple modes, and the preset parameters in the collected turning process are trained through a neural network model to obtain a recognition model for recognizing each mode.

[0041] In some embodiments of the present invention, the neural network model adopts LSTM, and the preset parameters include: power, X feed speed, Z-axis feed speed, vibration in the X-axis direction of the spindle, vibration in the Z-axis direction of the spindle, vibration in the X-axis of the turret, and vibration in the Z-axis.

[0042] It should be noted that currently, pattern recognition technology is mainly used for fault diagnosis. By collecting signals (including parameters such as current, voltage, power, and vibration) during the operation of the equipment, and using neural network or deep learning algorithms to judge the real-time stage of the equipment. Once a possible fault mode appears in the equipment, the system makes a judgment and gives an early warning, so as to realize the preventive and predictive maintenance of the equipment, improve the service life of the equipment and improve the processing quality of products.

[0043] The recognition model can adopt the LSTM model. The LSTM algorithm is evolved from the Recurrent Neural Network (RNN). Different from ordinary neural networks, LSTM has special memory neurons. On the basis of RNN, by adding an input threshold, a forgetting threshold, and an output threshold, the self-recurrent weight is changed. Through this structural design, the integration scale of the model can be dynamically changed with time under the condition of unchanged parameters, thereby avoiding the problems of gradient explosion and gradient disappearance.

[0044] The LSTM neural network structure is as Figure 2 shown, Figure 2 where f represents new memory, I is the input gate, F is the forgetting gate, and O is the output gate. Represents the calculation method of multiplying the corresponding elements in the matrix, C t Represents the memory data at the i-th step, x t Represents the model input data at the i-th step, Y t Represents the model output data at the i-th step. Self-loop represents the loop calculation, and Stated represents the update process of the memory data.

[0045] The training steps of the LSTM model are as Figure 3 shown, specifically including:

[0046] A1. Create a feature set and a target set. The creation method is as follows: After collecting the training data, use parameters such as power, vibration, feed, and coordinates as features, and use the corresponding processing mode of the data set as the target. Loop the above operations sequentially from the first piece of data to obtain the training feature set and the training target set;

[0047] A2. Initial setting of model parameters: Set the time step of the LSTM network to 40, the learning rate to 0.005, the number of hidden layers to 1, and the number of hidden neurons to 40. It can be understood that the above parameter settings are only an example of the present invention, and the corresponding parameter settings can be adjusted according to actual needs. In addition, the loss function uses the mean square error (MES), which is defined as follows:

[0048]

[0049] where: N is the amount of data in the data set; y i is the measured value corresponding to the output value of the LSTM model; is the output value of the LSTM model;

[0050] A3. Training of the LSTM model: Use the training feature set as the input of the LSTM model, use the output of the LSTM model as the prediction result of the training target set, use the prediction result of the training target set and the training target set itself as the input value of the LSTM model loss function to calculate the loss value in the LSTM model training process. Regard the above process as one epoch, and repeat the training for several epochs until the loss value in the LSTM model training process is lower than 0.0001 to obtain the optimal LSTM model under this data set.

[0051] S200. Compare the processing parameters of different workpieces under the same mode to obtain the first optimization criterion, and compare the processing parameters of the same workpiece under different modes to obtain the second optimization criterion;

[0052] According to some embodiments of the present invention, in step S200, comparing the processing parameters of different workpieces under the same mode to obtain the first optimization criterion includes:

[0053] S211. Statistically analyze the machining durations of workpieces with different preset quantities in the same mode, and determine a duration threshold;

[0054] S212. Adjust and optimize the feed rate in the corresponding mode based on the duration threshold.

[0055] For example, in the same machining stage (mode), it takes 4.6 s to machine some components, while it only takes 4.2 s to machine most parts. The difference in the required time in the middle is caused by the individual differences of the workpieces. If the load and vibration of the 0.4 s difference are not significant, the feed rate in the previous 0.4 s can be increased.

[0056] In some embodiments of the present invention, in step S200, comparing the machining parameters in different modes to obtain a second optimization criterion includes:

[0057] Compare the vibration parameters of the same workpiece in different modes. If the difference in the effective vibration values of the same workpiece in different modes exceeds a first threshold, adjust and optimize the feed rate in the corresponding mode.

[0058] For example, between different modes, by comparing the effective vibration values (including peak values and mean values), if the difference is above a set value, an algorithm is used to remind that there is room for optimization in the mode with a lower effective vibration value, and it is recommended to increase the feed rate to optimize the machining process.

[0059] According to some embodiments of the present invention, adjusting and optimizing the feed rate in the corresponding mode includes:

[0060] Increase the feed rate in the corresponding mode with a lower effective vibration value; and / or decrease the feed rate in the corresponding mode higher than a second threshold.

[0061] It should be noted that the present invention can either increase the feed rate in the corresponding mode with a lower effective value to optimize the turning process; or decrease the feed rate in the corresponding mode higher than the second threshold and increase the feed rate in the corresponding mode with a lower effective value, thereby improving the machining efficiency of the workpiece and increasing the tool life at the same time.

[0062] Among them, the "first threshold" can be understood as the set value of the difference between the effective vibration values (such as peak values or mean values) of the same workpiece in different machining modes, and the "second threshold" can be understood as the set value of the absolute value of the vibration of the same workpiece in different machining modes. It should be noted that since the set values of the absolute values of the vibrations of the same workpiece in different machining modes may be different, the "second threshold" can be multiple different values.

[0063] S300. Optimize the turning process based on the first optimization criterion and the second optimization criterion.

[0064] It should be noted that the current adaptive machining technology is mainly an intelligent control system that measures the spindle load in real time and adjusts the feed rate to improve machining efficiency. Lathes generally perform cutting according to the feed rate given in the lathe program. During the cutting process, the external machining conditions will change, but the equipment cannot make corresponding adjustments to the changing machining conditions at any time.

[0065] The present invention can fully consider the changes in external machining conditions caused by machining tolerances such as the hardness and stress of different workpieces, and optimize and adjust the turning process, thereby improving the rationality of the optimization of the turning process.

[0066] The turning process optimization method based on pattern recognition proposed by the present invention divides the machining process into different stages (patterns) by studying the specific machining process, collects relevant characteristic variables for training based on LSTM at the same time, and applies the trained model to the real-time collected machining process. By comparing the same machining stage (pattern) and different machining stages (patterns), once an abnormal / potential optimization point is detected, the system will give a prompt and optimization suggestions. Through the method proposed by the present invention, the characteristics of different machining processes can be taken into account, and the turning process can be systematically optimized, which can better improve the turning efficiency.

[0067] According to the machining optimization method for mass-produced mechanical parts according to an embodiment of the present invention, the machining optimization method uses the turning process optimization method based on pattern recognition as described above to optimize the machining of the cross shaft.

[0068] According to some embodiments of the present invention, the patterns divided in the turning process of mechanical parts include: rapid feed, rough turning of the end face, finish turning of the end face, rapid traverse, rough turning of the outer circle and drilling, rough turning of the remaining outer circle, finish turning of the outer circle, deceleration and loading / unloading.

[0069] For example, the turning process optimization method based on pattern recognition described above in the present invention can be used for the machining of the cross shaft, so as to optimize the turning process of the cross shaft. It can be understood that the above division of the cross shaft machining process is only an example for the convenience of understanding the solution of the present invention, and the cross shaft turning process can be different machining processes. For example, there is only one end face turning, without distinguishing between rough turning and finish turning, or there are other hole machining requirements, etc. It can be understood that the method provided by the present invention is not only applicable to the machining of cross shafts, but also can be applied to the machining of other shaft parts and similar bearing parts.

[0070] In some embodiments of the present invention, the turning machining parameters are obtained by adding a power sensor, a vibration sensor and the PLC of the machining equipment.

[0071] According to some embodiments of the present invention, a machining method optimizes a machining process of a mechanical part in real time based on machining parameters collected in real time.

[0072] The present invention uses an adaptive algorithm to calculate a reasonable feed speed as quickly as possible through the real-time collection of turning parameters, and timely adjusts the feed ratio. Under light load conditions, the feed speed is increased; under heavy load conditions, the feed speed is reduced, so that constant power cutting can be achieved and turning efficiency can be improved.

[0073] The following is a detailed description of the lathe process optimization method based on pattern recognition for lathe processing of heavy cross shafts according to the present invention with reference to the accompanying drawings. It is worth noting that the following description is only an exemplary description and should not be construed as a specific limitation of the present invention. For example, the lathe process optimization method of the present invention can also be applied to the lathe processing of medium-sized cross shafts, bearings, straight shafts and other parts produced in batches.

[0074] The efficiency of turning directly affects the processing cost and the market competitiveness of the enterprise. Therefore, the optimization of process parameters in the turning process has always been a problem that has troubled many parts processing companies. Traditional process parameter optimization is highly dependent on the experience of process personnel, and is generally a point-like, non-systematic optimization. There are also adaptive optimization products on the market, but such products generally depend on real-time load (power) without considering factors such as different vibrations and feeds, nor the characteristics of different processing stages, so the efficiency improvement is generally small.

[0075] The embodiment of the present invention divides the heavy-duty cross-axle lathe processing process into several stages according to the process characteristics, collects power, vibration, feed and other data in real time according to the process characteristics, and realizes pattern recognition at different stages of the processing process through the long short-term memory network (LSTM). By comparing the beats, power and vibration of the same pattern in different processed parts, the heavy-duty cross-axle lathe processing beat is optimized. The specific steps are as follows: Figure 4 As shown:

[0076] A1, such as Figure 4 As shown, the different stages of the turning process are defined: the power curve of the heavy cross-axis turning process and the different modes represented by different curves. By corresponding the process and the power curve, the power of a processing cycle is divided into 9 different modes. The meanings of different modes are shown in the following table.

[0077] Serial number Mode Remarks 1 Mode 1 Fast forward 2 Mode 2 Rough turning of end face 3 Mode 3 Finish turning of end face 4 Mode 4 Rapid feed 5 Mode 5 Rough turning of outer circle and drilling 6 Mode 6 Rough turning of remaining outer circle 7 Mode 7 Finish turning of outer circle 8 Mode 8 Deceleration 9 Mode 9 Loading and unloading

[0078] A2, according to the process characteristics, power and vibration sensors are installed to collect spindle feed, coordinate position and other parameters from the equipment PLC;

[0079] A3. The collected data are tagged (processing stage (mode)), and then input into the LSTM model for training (the input parameters are power, spindle X feed rate, vibration, Z-axis feed rate, vibration, turret X-axis vibration, and Z-axis vibration).

[0080] A4. Use the trained model to perform pattern recognition on the real-time machining process, as Figure 6 shown.

[0081] A5. Comparison of machining parameters for the same / different patterns:

[0082] 1) In the same machining stage (mode), it takes 4.6 s to machine some parts, while it only takes 4.2 s to machine most parts. The difference in the required time in the middle is caused by the individual differences of the workpieces. If the load and vibration of the 0.4 s difference are not significant, the feed rate in the previous 0.4 s can be increased.

[0083] 2) Between different patterns (referring to specific machining stages, such as rough turning of the outer circle, finish turning of the outer circle, rough turning of the end face, and finish turning of the end face), by comparing the effective values of vibration (including peak value and mean value), if the difference is above the set value, the algorithm is used to remind that there is room for optimization in the pattern with a lower effective value of vibration.

[0084] A6. According to the results of step A5, if either of the above two situations is found, the system prompts the potential optimization space and gives optimization suggestions.

[0085] In summary, through the method of the present invention, an optimization method based on pattern recognition for the machining process of heavy cross shafts, characteristic values for pattern recognition in the machining process of heavy cross shafts are proposed in the embodiments, and a system is established, which can detect the possibility of optimizing process parameters according to real-time measurement values and provide optimization suggestions.

[0086] Through the description of the specific implementation manners, it should be possible to understand more deeply and specifically the technical means and effects adopted by the present invention to achieve the predetermined purpose. However, the attached drawings are only for reference and illustration, and are not used to limit the present invention.

Claims

1. An optimization method for turning processes based on pattern recognition, characterized in that, Including: S100, collecting the machining parameters of the lathe, and judging the mode of the lathe machining through a pre-trained recognition model; S200, comparing the machining parameters of different workpieces in the same mode to obtain the first optimization criterion, and comparing the machining parameters of the same workpiece in different modes to obtain the second optimization criterion; Comparing the machining parameters of different workpieces in the same mode to obtain the first optimization criterion, including: S211, statistically analyzing the machining time of a preset number of different workpieces in the same mode, and determining the time threshold; S212, adjusting and optimizing the feed speed in the corresponding mode based on the time threshold; Comparing the machining parameters of the same workpiece in different modes to obtain the second optimization criterion, including: Comparing the vibration parameters of the same workpiece in different modes, if the difference in the effective vibration values of the same workpiece in different modes exceeds the first threshold, adjusting and optimizing the feed speed in the corresponding mode; S300, optimizing the lathe machining process based on the first optimization criterion and the second optimization criterion.

2. The optimization method for the turning process based on pattern recognition according to claim 1, wherein The method further includes: Pre-dividing the lathe machining process into multiple modes, and training through a neural network model with the preset parameters in the collected lathe machining process to obtain the recognition model for identifying each mode.

3. The method for optimizing the turning process based on pattern recognition according to claim 2, wherein The neural network model adopts LSTM, and the preset parameters include: power, feed speed in the X-axis direction, feed speed in the Z-axis direction, vibration in the X-axis direction of the main shaft, vibration in the Z-axis direction of the main shaft, vibration in the X-axis direction of the turret, vibration in the Z-axis direction.

4. The method for optimizing the turning process based on pattern recognition according to claim 1, wherein Adjusting and optimizing the feed speed in the corresponding mode includes: Increasing the feed speed in the corresponding mode with a lower effective vibration value; and / or decreasing the feed speed in the corresponding mode higher than the second threshold.

5. A processing optimization method for mass-produced mechanical parts, characterized in that, The machining optimization method adopts the machining optimization of mechanical parts in mass production by using the lathe machining process optimization method based on pattern recognition according to any one of claims 1-4.

6. The processing optimization method for mass-produced mechanical parts according to claim 5, characterized in that Obtaining the lathe machining parameters through the additionally installed power sensor, vibration sensor and the PLC of the processing equipment.

7. The machining optimization method for mass-produced mechanical parts according to claim 6, characterized in that, The machining optimization method is based on the real-time collected lathe machining parameters and performs real-time optimization on the lathe machining process of mechanical parts.

Images