Machine tool precision out-of-tolerance fault diagnosis method based on fusion of mechanism and deep learning

Abstract

The invention discloses a mechanism and deep learning fused machine tool precision out-of-tolerance fault diagnosis method, and belongs to the technical field of precision numerical control machine tool health management. Comprising the following steps: S10, constructing a tool nose nonlinear dynamic model of a nonlinear stiffness item K (x) considering thermal deformation influence; s20, the inherent frequency, damping and Taylor empirical formula coefficients of the whole machine are obtained in combination with the hammering excitation test and the cutting test; s30, the displacement change of a tool nose is obtained; s40, constructing an out-of-tolerance recognition library; and S50, calculating a precision index of the X (t) obtained under different working conditions, comparing the precision index with a threshold value to generate a qualified / out-of-tolerance label, and carrying out fault prediction based on a 1D-CNN-BiLSTM fusion network training classification model. And finally verifying the precision out-of-tolerance problem of the machining process in combination with field experimental data. The method can solve the problems that the traditional diagnosis technology is difficult to capture in real time, the precision is out-of-tolerance, and generalization is insufficient due to separation of a physical model and a data driving method.

Description

Technical Field

[0001] This invention relates to the field of health management technology for precision manufacturing equipment, and in particular to a real-time diagnostic method for machine tool machining accuracy deviations that integrates nonlinear dynamic modeling of the tool tip, multi-domain signal analysis, and deep learning. Background Technology

[0002] As the manufacturing industry moves towards high-speed and high-precision machining, out-of-tolerance failures, as a core failure mode in the operation of high-precision machine tools, are crucial for ensuring machining accuracy, extending equipment lifespan, and improving manufacturing quality. Real-time, accurate diagnosis and prediction of out-of-tolerance failures are key to achieving proactive control of the machining process, dynamic optimization of process parameters, and predictive maintenance, which is of great significance for ensuring the reliability and economy of the production line.

[0003] Currently, diagnostic methods for machine tool accuracy deviations can be broadly categorized into two types: physical model-based methods and data-driven methods. Physical model-based methods typically predict deformation or vibration by establishing thermodynamic or dynamic models of the machine tool structure. For example, this involves finite element analysis of the temperature and thermal stress fields of the entire machine tool or key components, or establishing linear vibration equations for the tool tip. While these methods have clear physical mechanisms, they have significant limitations: First, a real machine tool is a complex nonlinear system whose stiffness, damping, and other parameters change with temperature, displacement, and operating conditions. Simplified linear models struggle to accurately describe its true dynamic behavior, leading to insufficient prediction accuracy. Second, high-fidelity physical models are complex to construct, computationally expensive, and heavily reliant on system parameters that are difficult to obtain accurately, posing a significant challenge for real-time applications in actual industrial settings. Data-driven methods, on the other hand, bypass complex physical mechanisms and directly extract fault characteristics from sensor data (such as vibration, temperature, and acoustic emission signals) using machine learning algorithms (such as support vector machines and shallow neural networks). While these methods reduce reliance on prior knowledge to some extent, they also have inherent drawbacks. Firstly, their performance heavily depends on massive amounts of high-quality labeled data covering all fault modes. However, in actual industrial applications, especially for progressive faults like those with out-of-tolerance precision, obtaining a complete fault sample library is extremely difficult. Secondly, purely "black-box" data-driven models lack physical interpretability; their decision-making logic is disconnected from the actual physical processes of the machine tool. When operating conditions change in ways not covered by the training data, the model's generalization ability and reliability drop sharply. In recent years, although some studies have attempted to combine these two methods simply, they are mostly loosely coupled and fail to achieve deep integration of mechanisms and data. For example, CN120354555A and CN120597697A only use the output of the physical model as a feature of the data-driven model, failing to fundamentally solve the problem of synergistic development between accurate description of nonlinear dynamic behavior and adaptive extraction of deep features. Therefore, existing technologies struggle to achieve real-time, robust diagnosis of machine tool out-of-tolerance precision faults, especially early and minor faults, while ensuring the physical consistency of the model. In summary, existing technologies suffer from several core problems, including insufficient accuracy due to simplified physical models, poor generalization of data-driven methods and reliance on large amounts of data, and a lack of deep integration between mechanisms and data models. This invention aims to overcome these shortcomings by proposing a novel method that deeply integrates nonlinear dynamic mechanism modeling with deep learning-based intelligent diagnosis. Summary of the Invention

[0004] The purpose of this invention is to construct a nonlinear dynamic model of the whole machine under the influence of thermal deformation stiffness, and to build a precision deviation identification library by combining the mathematical mapping relationship between tool tip displacement and machining accuracy, so as to solve the problems of insufficient accuracy of physical models, reliance on a large amount of data for data-driven approaches, and high experimental costs and difficulty in application.

[0005] This invention provides a nonlinear dynamic model of the entire machine that considers the effects of thermal deformation. It constructs a precision deviation identification database by combining the mathematical mapping between tool tip displacement and machining accuracy, and performs precision deviation fault identification. Specifically, it includes the following steps:

[0006] First, the change in overall machine stiffness under the influence of thermal deformation must be calculated. Thermal deformation can be obtained through mutual verification between thermo-mechanical coupling simulation and actual testing of the entire machine. The heat sources in the overall machine simulation analysis include the guide rail slider pair, bearing pairs such as the main shaft, the lead screw and nut pair, and the motor. The specific heat generation formula for these heat sources is:

[0007]

[0008] in, It is the total frictional torque of the bearing. It is the bearing speed. It is the frictional torque of the lead screw and nut. It refers to the rotational speed of the ball screw. It is the coefficient of kinetic friction. It is a load perpendicular to the friction surface of the guide rail. It is the sliding speed. It is the mechanical equivalent of heat. It refers to the efficiency of the motor.

[0009] The central boundary condition is the natural convection coefficient between the entire machine and the air. The forced convection coefficient between the spindle and other bearings during high-speed rotation and air. The forced convection coefficient of the lead screw and nut pair when rotating with air The forced convection coefficient of motor air cooling Its specific functional relationship is...

[0010]

[0011] in, It is the area perpendicular to the direction of heat flow. It is the temperature difference on the surface of the object.

[0012] Secondly, the nonlinear dynamic model of the whole machine under the influence of thermal deformation stiffness is constructed as follows:

[0013]

[0014] Where k eq (t) Total dynamic stiffness, The inherent frequency of the whole machine Overall damping ratio, Varying cutting force. Where k eq(t) is represented as:

[0015]

[0016] Where k s k c0 For structural stiffness and cutting stiffness at room temperature, The value represents the thermal deformation over time. This is the nominal cutting depth.

[0017] Then, by combining hammer excitation experiments and cutting experiments, the parameters in the above formula are obtained, and the tool tip displacement curve under the influence of thermal deformation is solved using the variable step size Runge-Kutta method. Furthermore, a fault database for accuracy deviations is constructed by combining the mathematical mapping relationship between tool tip displacement and machining accuracy, using the original signals. It may contain noise and non-steady-state components, so preprocessing and obtaining its low-pass and band-pass filters are necessary. Meanwhile, because... Different frequency components correspond to different categories of processing errors. The specific mathematical mapping model is as follows:

[0018] The low-pass filter component is related to dimensional accuracy:

[0019]

[0020] Where G is the geometric coefficient. for low-pass filter function

[0021] The bandpass filter component is related to shape accuracy:

[0022]

[0023] in for The bandpass filter function (within one rotation period)

[0024] Based on its mathematical relationships, it can be seen that the Runge-Kutta method can be used to solve the problem. This is directly converted into quantifiable accuracy metrics. By setting corresponding accuracy thresholds, error prediction is achieved when the aforementioned components exceed the threshold. Furthermore, it provides corresponding supervised learning labels for subsequent machine learning, allowing the results obtained under different operating conditions (different speeds, feed rates, and thermal deformation parameters) to be used. Accuracy indicators were calculated and compared with thresholds to generate "qualified / out-of-tolerance" labels, which were then used to train a classification model for fault prediction. Finally, the accuracy deviation problem in the processing process was verified using field experimental data.

[0025] This invention provides a machine tool precision deviation fault diagnosis method based on nonlinear dynamic modeling under the influence of thermal stiffness and deep learning fusion. Its purpose is to solve the problems of traditional diagnostic technology's inability to capture precision deviations in real time and the lack of generalization caused by the separation of physical models and data-driven methods, thus laying a solid foundation for predictive maintenance and intelligent process optimization of machine tools.

[0026] The above-mentioned technical objective of this disclosure is achieved through the following technical solution:

[0027] The beneficial effects of this disclosure are as follows: The machine tool precision deviation fault diagnosis method based on the fusion of nonlinear dynamic modeling and deep learning, as described in this invention, is based on a high-fidelity nonlinear dynamic model. It comprehensively considers the influence of time-varying stiffness due to thermal deformation and cutting excitation on tool tip vibration, accurately characterizing the dynamic behavior of the machine tool machining process. Furthermore, a high-precision fault classification model is constructed based on a 1D-CNN-BiLSTM fused deep learning network. This method encompasses a complete technical chain, from constructing a nonlinear dynamic model, acquiring physical parameters, analyzing vibration response, establishing a deviation identification library, to training an intelligent classification model, forming a scientific diagnostic system driven by both mechanism and data. Finally, field experiments verified the effectiveness of this method in real-time capture of precision deviations and predictive maintenance, providing reliable technical support for the proactive control of machine tool machining precision and dynamic optimization of process parameters. Attached Figure Description

[0028] Figure 1 This is a flowchart of a machine tool accuracy deviation fault diagnosis method based on the fusion of nonlinear dynamics modeling and deep learning.

[0029] Figure 2 The diagram shows the displacement of the tool tip under the influence of thermal deformation.

[0030] Figure 3 This is a flowchart of a fault identification process based on a 1D-CNN-BiLSTM fusion deep learning network to identify faults with excessive accuracy.

[0031] Figure 4 This is a diagram for identifying machine tool precision deviation faults. Detailed Implementation

[0032] The following is combined with Figure 1 The flowchart shown below illustrates a machine tool accuracy deviation fault diagnosis method based on the fusion of nonlinear dynamics modeling and deep learning, providing a detailed description of the technical solutions in this embodiment of the invention. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are all within the scope of protection of this invention.

[0033] like Figure 1 As shown, this disclosure proposes a machine tool precision deviation fault diagnosis method based on the fusion of nonlinear dynamic modeling and deep learning. This method deeply integrates physical mechanisms and data-driven intelligence on the basis of a high-fidelity nonlinear dynamic model to achieve accurate prediction and diagnosis of machining precision deviation faults. First, a nonlinear dynamic model of the tool tip considering the influence of time-varying stiffness due to thermal deformation is constructed to accurately characterize the dynamic behavior of the machine tool machining system. Then, the overall dynamic parameters and cutting force coefficients are obtained through hammer excitation tests and cutting tests to ensure the physical authenticity of the model parameters. Subsequently, the tool tip vibration displacement response is obtained through numerical analysis using the variable-step-size Runge-Kutta method, and a deviation identification database is constructed based on its strict mathematical mapping relationship with machining precision. Finally, a 1D-CNN-BiLSTM fused deep learning network is used to perform feature learning and pattern recognition on the vibration response under different working conditions, training a high-precision fault classification model. The method's effectiveness in real-time capture of precision deviations and predictive maintenance is verified by field experimental data, providing reliable technical support for the active control of machine tool machining precision and dynamic optimization of process parameters.

[0034] Figure 2 The diagram shows the tool tip displacement under the influence of thermal deformation, such as... Figure 2 As shown, this analysis is based on a nonlinear dynamic model of the entire machine, and the Runge-Kutta method is used to numerically solve the dynamic equations, where the thermal deformation effect is introduced through a time-varying stiffness term. The case without thermal effects (dashed line) only considers the basic cutting force disturbance, while the case with thermal effects (solid line) adds a stiffness attenuation term caused by thermal deformation, resulting in displacement exhibiting a reference drift and increased amplitude. Furthermore, the figure shows that the curve with thermal effects deviates significantly from the curve without thermal effects over time.

[0035] Figure 3 This is a flowchart of a precision error identification process based on a 1D-CNN-BiLSTM fusion deep learning network. First, various operating conditions of the machine tool are set, such as spindle speed (n), feed rate (f), and depth of cut (a0), as inputs to the physical model. Next, the simulated tool tip displacement data x(t) undergoes signal processing, and its corresponding machining state is automatically determined based on physical laws. If any error exceeds the limit, this x(t) data is marked as "out of tolerance"; otherwise, it is marked as "acceptable." Finally, the dataset constructed in the previous step is input into the deep learning network for training. After training, the optimal model parameters (such as weights and biases) are saved for direct use in the online diagnostic stage to complete the precision error identification.

[0036] Figure 4This demonstrates that extracting the low-frequency components of the displacement signal through moving average filtering achieves a quantitative assessment of dimensional accuracy. When the filtered displacement value exceeds the set dimensional accuracy threshold (8 μm), the gray-filled area clearly indicates the timing and severity of the dimensional deviation fault, providing intuitive verification of the mathematical mapping relationship Δd = G·X_dc(t). Simultaneously, shape accuracy is assessed through vibration signal amplitude analysis, reflecting the peak-to-peak characteristics of the bandpass component of X_ac(t). When the vibration amplitude exceeds the shape accuracy threshold (5 μm), the marked area clearly shows the roundness error deviation, confirming that thermal deformation not only causes dimensional drift but also degrades the shape accuracy of the machined surface by altering the system's dynamic characteristics.

Claims

1. A method for diagnosing machine tool precision deviation faults by integrating mechanism and deep learning, characterized in that, Includes the following steps: Step 1: Calculate the heat source load and boundary conditions of the entire machine; Step 2: Construct a nonlinear dynamic model of the entire machine under the influence of thermal deformation. The model equates time-varying thermal deformation to a time-varying stiffness term, and its dynamic equation is expressed as: ， Where m and c are the equivalent mass and damping of the system, respectively, x(t) is the tool tip displacement response, F(t) is the cutting excitation force, and k eq (t) represents the total dynamic stiffness, and its expression is: ， Where, k s For structural stiffness, k c0 Here, a is the cutting stiffness at room temperature, a0 is the nominal depth of cut, and δ is the cutting stiffness at room temperature. T (t) represents the thermal deformation value as a function of time; Step 3: Obtain model parameters and solve for the tool tip displacement response. Specifically, obtain the natural frequency of the entire machine through a hammer impact excitation experiment. With the damping ratio ξ, the cutting force coefficient is obtained through cutting experiments, and the nonlinear dynamic equation described in step 2 is numerically solved using the variable step size Runge-Kutta method to obtain the time-domain signal x(t) of the tool tip displacement. Step 3.1: Construct a precision deviation identification library based on physical mapping relationships; perform signal processing on the tool tip displacement signal x(t) obtained in Step 3, extract its low-pass filtered component Xdc(t) and band-pass filtered component Xac(t), and convert them into precision indicators according to the following mathematical mapping relationship: dimensional accuracy error: Shape accuracy error: ; Where G is the geometric coefficient; by setting the dimensional accuracy threshold θ d and shape accuracy threshold θ f The calculated accuracy index is compared with the threshold, and the supervised learning label of "qualified" or "out of tolerance" is automatically generated, thereby constructing an accuracy deviation identification library containing multi-condition data. Step 4: Accuracy deviation fault identification based on deep learning network; The tool tip displacement data x(t) and its corresponding labels in the accuracy deviation identification library constructed in Step 3 are used as the training set and input into a deep learning model that integrates a one-dimensional convolutional neural network and a bidirectional long short-term memory network for training to obtain an accuracy deviation fault classification model. Step 5: Real-time diagnosis and application verification; The real-time tool tip displacement signal obtained from online monitoring is input into the fault classification model trained in Step 4 to realize real-time diagnosis and prediction of machining accuracy deviation faults, and the accuracy of the diagnosis results is verified through on-site machining experiments.

2. The machine tool precision deviation fault diagnosis method integrating mechanism and deep learning according to claim 1, characterized in that, The thermal deformation value δ mentioned in step 1 T (t), the heat sources involved in its calculation include the guide rail slider pair, the main shaft bearing pair, the lead screw nut pair, and the motor. The heat generation rate of the heat source is calculated using the following formula: ； in, It is the total frictional torque of the bearing. It is the bearing speed. It is the frictional torque of the lead screw and nut. It refers to the rotational speed of the ball screw. It is the coefficient of kinetic friction. It is a load perpendicular to the friction surface of the guide rail. It is the sliding speed. It is the mechanical equivalent of heat. It refers to the efficiency of the motor.

3. The machine tool precision deviation fault diagnosis method integrating mechanism and deep learning according to claim 1, characterized in that, The boundary conditions in step 1 are the natural convection coefficient h1 between the whole machine and the air, the forced convection coefficient h2 between the bearing and the air during high-speed rotation, the forced convection coefficient h3 between the lead screw and nut pair and the air during rotation, and the forced convection coefficient h4 between the machine and the air-cooled motor; the functional relationships are as follows: in, It is the area perpendicular to the direction of heat flow. It is the temperature difference on the surface of the object.

4. The machine tool precision deviation fault diagnosis method integrating mechanism and deep learning according to claim 1, characterized in that, The low-pass filter component Xdc(t) mentioned in step 3 is used to evaluate the dimensional accuracy of the workpiece, and its filter cutoff frequency corresponds to the feed motion frequency of the machine tool; the band-pass filter component Xac(t) is used to evaluate the shape accuracy of the workpiece, and its passband frequency corresponds to the spindle rotation frequency.

5. The machine tool precision deviation fault diagnosis method integrating mechanism and deep learning according to claim 1, characterized in that, The automatic generation of supervised learning labels mentioned in step 3.1 is specifically implemented as follows: for any tool tip displacement signal x(t) under any working condition, if either Δd>θd or Δf>θf is satisfied, it is marked as "out of tolerance"; otherwise, it is marked as "qualified".

6. The machine tool precision deviation fault diagnosis method integrating mechanism and deep learning according to claim 1, characterized in that, The one-dimensional convolutional neural network mentioned in step 4 is used to adaptively extract local multi-scale fault features from the blade tip displacement signal, and the bidirectional long short-term memory network is used to capture long-range temporal dependencies in the signal.

7. The machine tool precision deviation fault diagnosis method integrating mechanism and deep learning according to claim 1, characterized in that, The model parameters obtained through experiments in step 3 are combined with the physical model constructed in step 2 to form a diagnostic system driven by both mechanism and data, ensuring the interpretability and generalization ability of the model.

Images