A tool wear prediction method for machine tool vibration time domain signal imaging processing

Abstract

The application discloses a tool wear prediction method for machine tool vibration time domain signal imaging processing, and relates to a predictive maintenance system, which comprises the following steps: collecting time domain signal data of machine tool vibration; converting the time domain signal into a frequency domain signal of a sinusoidal wave with different frequencies through Fourier transform; analyzing the sinusoidal wave frequency domain signal to obtain corresponding vibration amplitudes and phases, converting the sinusoidal wave frequency domain signal into vibration image points, and then converting vibration image point data at the same time into a vibration matrix to form a vibration image with complete machine tool vibration information at the same time; performing feature extraction on the vibration image through a convolutional neural network, and simultaneously performing model training on the convolutional neural network through an improved POWELL algorithm; and outputting the detection result to obtain tool wear data and perform predictive analysis.

Description

Technical Field

[0001] This invention relates to the field of predictive maintenance systems, and in particular to a method for predicting tool wear through time-domain image processing of machine tool vibration signals. Background Technology

[0002] In the machining process of parts using CNC machine tool spindle cutters, the degree of tool wear is a key factor affecting the yield of finished parts. Currently, enterprises mainly rely on manual experience to solve problems such as tool wear, breakage, and tool breakage during production. However, this approach has problems because enterprises cannot accurately predict when equipment failures will occur, especially spindle tool wear or breakage, leading to significant losses. Therefore, there is an urgent need to develop a predictive maintenance system for machine tool wear to predict tool life and enable timely maintenance.

[0003] Chinese patent CN202011076025.1 discloses a method for predicting the wear state of CNC machine tool tools based on parallel deep neural networks. This method involves installing force gauges, accelerometers, and acoustic sensors on the CNC machine tool's worktable and fixture; conducting milling experiments to collect cutting force, vibration, and acoustic signals during the milling process, obtaining multi-sensor data, and collecting tool wear data; preprocessing to obtain training data and test data; establishing a parallel deep neural network model; inputting the processed training data and tool wear labels into the parallel deep neural network for offline training; and inputting the test multi-sensor data into the trained model to predict tool wear online in real time. However, this method suffers from the problem of sequentially inputting the same dataset into the network model for each training iteration, which can easily lead to overfitting and slow convergence. To improve computational efficiency, real-time performance is often sacrificed. Summary of the Invention

[0004] To address the problems of overfitting and slow convergence speed in existing machine tool wear prediction systems, this invention provides a tool wear prediction method using image processing of machine tool vibration time-domain signals. This method simplifies the problem of finding the extrema of a multivariate function into a one-dimensional extremum problem by eliminating the need to calculate function gradients and the concept of algorithm application direction sets, thus accelerating function convergence. The specific technical solution is as follows:

[0005] The present invention provides a method for predicting tool wear by image processing of machine tool vibration time-domain signals, specifically as follows: acquiring time-domain signal data of machine tool vibration;

[0006] The time-domain signal is converted into a sinusoidal frequency-domain signal with different frequencies by using Fourier transform;

[0007] The sinusoidal frequency domain signal is analyzed to obtain the corresponding vibration amplitude and phase. The sinusoidal frequency domain signal is converted into vibration image points. Then, the vibration image point data at the same moment is converted into a vibration matrix to form a vibration image with complete vibration information of the machine tool at the same moment.

[0008] Feature extraction of vibration images is performed using a convolutional neural network, while the convolutional neural network model is trained using an improved POWELL algorithm.

[0009] The vibration image processed by the convolutional neural network is converted into tool wear detection results for output. Based on the tool wear detection results, tool wear data is obtained and predictive analysis is performed.

[0010] The machine tool vibration time-domain signal data is collected, including: acquiring the tool vibration signal through a three-axis accelerometer installed at a fixed position on the turret; after the three-axis accelerometer acquires the charge signal, it is filtered and amplified by a filter amplifier to output a sinusoidal frequency domain signal.

[0011] The sinusoidal voltage signal is processed by the multi-condition holographic intelligent signal processor AIC8660-900 and then transmitted to the host software via TCP / IP communication. The host software converts the time-domain signal into sinusoidal frequency-domain signals of different frequencies for analysis. By analyzing the sinusoidal frequency-domain signals, the corresponding vibration amplitude and frequency are obtained, and the sinusoidal frequency-domain signals are further converted into image signals.

[0012] The convolutional neural network normalizes the vibration images.

[0013] Feature extraction of vibration images by convolutional neural networks includes: horizontal or vertical flipping of vibration images, random angle rotation of vibration images, proportional scaling of vibration images, random cropping of vibration images, displacement of vibration images, addition of Gaussian noise to vibration images, and color enhancement of vibration images.

[0014] The convolutional neural network extracts features from the input data by sliding the convolutional kernel across the data to be extracted with a set stride. The pooling layer compresses the feature data values ​​to reduce the dimensionality of the data.

[0015] Improvements to the POWELL algorithm for training convolutional neural network models include:

[0016] Construct mutually conjugate directions, including a one-dimensional search along one direction and the next round of search along another direction;

[0017] The optimization process is divided into several stages. Each stage, i.e. each iteration, consists of n+l one-dimensional searches. In each stage of the algorithm, the search is first performed along the known n directions to obtain a best point. Then, the search is performed along the direction of the line connecting the initial point of this stage and the best point to obtain the best point of this stage.

[0018] Then replace one of the first n directions with the last search direction to begin the next stage of iteration.

[0019] The specific construction process of the improved POWELL algorithm is as follows:

[0020] (1) Choose n distinct initial values. and Neither parallel z 1 The allowable error is ω > 0, i = 1, k = 1;

[0021] (2) Calculation make

[0022] (3) If i < k-1, then let i := i+1 and go to (2); if i = k-1, then let Go to (2); if this simplified form i = k, then go to (4);

[0023] (4) If k = n, then go to (5); if k < n, then let k = k + 1, i = i + 1, and go to (2);

[0024] (5) If i = 1 and k = n, then stop. Otherwise, go to (6);

[0025] (6) Calculation

[0026] (7) If f0-2f n +f * ≥2Δ, then Turn (2), otherwise turn (8);

[0027] (8) Let z i =z i+1 (i = m ~ n-1), calculate make i=1,k=n (2).

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] 1. This invention reduces the influence of other factors on the data during long-term machine tool vibration measurement by recording data in an incremental manner; at the same time, it also normalizes the data to reduce the impact caused by excessive differences in data dimensionality.

[0030] 2. This invention randomly shuffles the order of the training dataset to avoid feeding the network model with the same dataset order each time it is trained, which greatly alleviates the overfitting of the data and the loss of authenticity and accuracy.

[0031] 3. This invention trains on multiple different copies of each image, which is beneficial for generalizing to variations of the same object, thereby improving the model's generalization ability and robustness, and improving the prediction effect of tool wear in small samples. Attached Figure Description

[0032] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below; in all the drawings, similar elements or parts are generally identified by similar reference numerals; in the drawings, the elements or parts are not necessarily drawn to actual scale.

[0033] Figure 1 This is a flowchart of the method of the present invention;

[0034] Figure 2 This is a diagram illustrating the normalization process for a convolutional neural network. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0036] It should be understood that, when used in this specification and the appended claims, the terms “comprising” and “including” indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0037] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention; as used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0038] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0039] The following reference Figure 1 and Figure 2 The present invention will be further explained and described.

[0040] The system collects time-domain vibration signal data of the machine tool; converts the time-domain signal into a sinusoidal frequency-domain signal with different frequencies using Fourier transform; analyzes the sinusoidal frequency-domain signal to obtain the corresponding vibration amplitude and phase, converts the sinusoidal frequency-domain signal into vibration image points, and then converts the vibration image point data at the same moment into a vibration matrix to form a vibration image with complete machine tool vibration information at the same moment; extracts features from the vibration image using a convolutional neural network, and trains the convolutional neural network model using an improved POWELL algorithm; the vibration image processed by the convolutional neural network is converted into tool wear detection results for output, and tool wear data is obtained and predictive analysis is performed based on the tool wear detection results.

[0041] In some embodiments, the vibration acquisition hardware mainly includes a general-purpose IEPE accelerometer, an IEPE (ICP) vibration signal sensor, and a multi-condition holographic intelligent signal processor (AIC8660-900) from Aerospace Intelligent Control Technology Co., Ltd. The predictive maintenance system uses a triaxial accelerometer to acquire tool vibration signals. Since the workpiece is moving during machining and the tool rotates during tool changes, the vibration sensor needs to be installed at a fixed position on the tool turret. The vibration signal acquired by the triaxial accelerometer is converted into a sinusoidal voltage signal output. Essentially, it acquires a charge signal, filters and amplifies it through a filter amplifier, and then outputs a voltage signal. After processing by the multi-condition holographic intelligent signal processor (AIC8660-900), the vibration information is transmitted to the host software via TCP / IP communication. Leveraging the powerful signal processing capabilities of the host software, the time-domain signal is simultaneously converted into sinusoidal frequency-domain signals of different frequencies for analysis. By analyzing the sinusoidal frequency-domain signals, the corresponding vibration amplitude and frequency are obtained, and the sinusoidal frequency-domain signals are further converted into image signals.

[0042] The vibration field of a CNC machine tool is the sum of the vibration distribution points within the machine tool at a given instant. It is a distribution function of time and spatial location points, and its vibration distribution belongs to an unsteady vibration field. Converting vibration field data into vibration image points preserves the complete information of the machine tool vibration signal; simultaneously, using deep learning modeling methods avoids nonlinearity and coupling problems between vibration measurement points. The conversion approach is based on the numerical attributes of the image, i.e., the image consists of values ​​from 0 to 256. Vibration measurement point data at the same instant are converted into a vibration matrix, thus forming a vibration image containing complete vibration information of the machine tool at the same instant.

[0043] Convolutional neural networks (CNNs) are widely used in image processing due to their powerful feature extraction capabilities for high-dimensional data. A typical CNN consists of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. For example... Figure 2 The diagram shows a step of a convolutional neural network. The convolutional kernel slides across the data to be extracted with a set stride to extract features from the input data. The pooling layer compresses the feature data values ​​to reduce the dimensionality of the data.

[0044] Multi-parameter optimization problems first require determining a measure based on the specific problem, then defining an appropriate objective function according to the requirements, and finally optimizing the objective function to obtain the result. The properties of the objective function are crucial. Ideally, the objective function should be a continuous, smooth, convex function. This allows for the use of classic optimization algorithms, such as the Gauss-Newton method, the LM algorithm, and the steepest descent method. If the objective function has poor properties and contains numerous local extrema, classic optimization algorithms will terminate at these local extrema, resulting in incorrect registration parameters. Modern optimization algorithms, such as simulated annealing and genetic algorithms, can overcome local extrema problems. However, these algorithms often have slow convergence speeds, sacrificing real-time performance for better results. The Powell optimization algorithm, which avoids calculating function gradients and applies the concept of direction sets, simplifies the problem of finding the extrema of a multivariate function into a one-dimensional problem, thus accelerating convergence. The Power11 algorithm first constructs conjugate directions (a one-dimensional search along one direction will not interfere with the next round of search along another direction; such directions are called conjugate directions). This method divides the optimization process into several stages, each stage (one iteration) consisting of n+1 one-dimensional searches. In each stage of the algorithm, it first searches sequentially along the known n directions to obtain a best point. Then, it searches along the direction connecting the initial point of this stage to the best point to find the best point of this stage. Finally, it replaces one of the first n directions with the final search direction and begins the next iteration. Its specific construction process is as follows:

[0045] (1) Choose n distinct initial values. and Neither parallel z 1 The allowable error is ω > 0, i = 1, k = 1;

[0046] (2) Calculation make

[0047] (3) If i < k-1, then let i := i+1 and go to (2); if i = k-1, then let Go to (2); if this simplified form i = k, then go to (4);

[0048] (4) If k = n, then go to (5); if k < n, then let k = k + 1, i = i + 1, and go to (2);

[0049] (5) If i = 1 and k = n, then stop. Otherwise, go to (6);

[0050] (6) Calculation

[0051] (7) If f0-2f n +f * ≥2Δ, then Turn (2), otherwise turn (8);

[0052] (8) Let z i =z i+1 (i = m ~ n-1), calculate make i=1,k=n, turn to (2), thus forming a round of data iteration.

[0053] Incremental data recording reduces the impact of other factors on the authenticity and accuracy of vibration signal data during long-term measurements. It also normalizes the data, mitigating the effects of large dimensionality differences. Randomly shuffling the order of the training dataset avoids feeding the same dataset into the network model for each training iteration. This helps alleviate overfitting to some extent. During model training, data augmentation techniques such as horizontal or vertical image flipping, random angle rotation, proportional scaling, random cropping, image displacement, adding Gaussian noise, and color enhancement are used to improve the accuracy and generalization of the error model. Training on multiple copies of each image helps generalize to variations of the same object, thus improving the model's generalization ability and robustness. Practice shows that the model performs exceptionally well in predicting tool wear with a small sample size, achieving an accuracy of 92%.

[0054] Those skilled in the art will recognize that the units described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0055] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0056] In addition, the functional units in the embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of software functional units.

[0057] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A tool wear prediction method for machine tool vibration time domain signal imaging processing, characterized in that: The method comprises the following steps: Collecting time domain signal data of machine tool vibration; Converting the time domain signal into a frequency domain signal of a sinusoidal wave with different frequencies through Fourier transform; Analyzing the sinusoidal wave frequency domain signal to obtain the corresponding vibration amplitude and phase, converting the sinusoidal wave frequency domain signal into a vibration image point, and then converting the vibration image point data at the same time into a vibration matrix to form a vibration image with complete vibration information of the machine tool at the same time; Extracting features of the vibration image through a convolutional neural network, and training the convolutional neural network through an improved POWELL algorithm; The vibration image processed by the convolutional neural network is converted into a tool wear detection result for output, and the tool wear data is obtained and analyzed for predictive analysis according to the tool wear detection result; The sinusoidal voltage signal is processed by a multi-working-condition holographic intelligent signal processor AIC8660-900, and then transmitted to the upper software through TCP / IP communication. The upper software converts the time domain signal into a frequency domain signal of a sinusoidal wave with different frequencies for analysis. The vibration amplitude and frequency are obtained by analyzing the sinusoidal wave frequency domain signal, and the sinusoidal wave frequency domain signal is further converted into an image signal The convolutional neural network extracts features of the input data by sliding on the data to be extracted with a set step size through a convolution kernel. The pooling layer compresses the feature data value and reduces the dimension of the data. The improved POWELL algorithm for training the convolutional neural network includes constructing mutually conjugate directions, including one-dimensional search along a certain direction and next round search along another direction; The optimization process is divided into several stages, and each stage, i.e. each iteration, consists of n+l one-dimensional searches. In each stage of the algorithm, a best point is obtained by searching along the known n directions in turn, and then searching along the direction of the line connecting the initial point of the stage and the best point to obtain the best point of the stage; The last search direction is used to replace one of the previous n directions, and the iteration of the next stage begins.

2. The tool wear prediction method of image processing of machine tool vibration time domain signal according to claim 1, characterized in that, Collecting time domain signal data of machine tool vibration, including: collecting tool vibration signal through a three-axis acceleration sensor installed at a fixed position on the tool turret. The three-axis acceleration sensor collects charge signals, and then outputs a sinusoidal wave frequency domain signal after filtering and operational amplification by a filter amplifier.

3. The tool wear prediction method of machine tool vibration time domain signal imaging processing according to claim 1, characterized in that: The convolutional neural network normalizes the vibration image.

4. The tool wear prediction method of machine tool vibration time domain signal imaging processing according to claim 1, characterized in that: The feature extraction of the vibration image by the convolutional neural network includes: horizontally or vertically flipping the vibration image, randomly rotating the vibration image, scaling the vibration image, randomly cropping the vibration image, shifting the vibration image, adding Gaussian noise to the vibration image, and enhancing the color of the vibration image.

Images