Numerical control machining tool health state online evaluation system and method based on multi-sensor fusion

Abstract

The invention discloses a numerical control machining tool health state online evaluation system and method based on multi-sensor fusion. Comprising a data acquisition module, a noise data preprocessing module, a noise feature analysis and mode recognition module, a tool health state evaluation module and a multi-source data fusion module, and the data acquisition module, the noise data preprocessing module, the noise feature analysis and mode recognition module and the tool health state evaluation module are connected in sequence. The data acquisition module and the cutter health state evaluation module are both connected with the multi-source data fusion module, and the multi-source data fusion module is connected with the numerical control system. By integrating various sensor data, noise interference of a single sensor is reduced, the accuracy and stability of cutter health state evaluation are improved, and intelligent online evaluation of the cutter health state is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of tool condition monitoring for numerically controlled machine tools, and particularly to an online evaluation system and method for the health status of numerically controlled machining tools based on multi-sensor fusion. Background Art

[0002] During the numerical control machining process, the health status of the tool directly affects the machining accuracy, surface quality, and production efficiency of the workpiece. If problems such as tool wear and breakage cannot be detected and processed in a timely manner, it will not only lead to a decline in machining quality but may also cause damage to the machine tool and increase production costs.

[0003] Currently, tool condition monitoring is mainly divided into direct methods and indirect methods. The direct method directly measures the position or shape change at the cutting edge of the tool in some way, thereby directly reflecting the wear state of the tool. However, this method is complex to install and implement and has a high cost, making it difficult to achieve real-time online measurement. The indirect method collects machining process signals related to tool wear through various sensors and performs corresponding processing and analysis to obtain the wear state of the tool. However, existing indirect monitoring methods usually use a certain characteristic parameter of a single sensor signal to represent the wear state of the tool. The accuracy of monitoring is limited by the accuracy of a single sensor, and the monitoring stability is poor, unable to effectively monitor the tool condition. At the same time, in current manufacturing, the preprocessing, feature extraction, and feature selection of sensor information mainly rely on the signal processing technology and diagnostic experience of technicians, far from meeting the requirements of intelligence.

[0004] Therefore, a stable and accurate intelligent online evaluation system and method for tool health status are needed. Summary of the Invention

[0005] Aiming at the deficiencies in the prior art, the purpose of the present invention is to provide an online evaluation system and method for the health status of numerically controlled machining tools based on multi-sensor fusion. By integrating multiple sensor data, it reduces the noise interference of a single sensor, improves the accuracy and stability of tool health status evaluation, and realizes the intelligent online evaluation of tool health status.

[0006] To achieve the above purpose, the present invention is implemented through the following technical solutions: An online evaluation system for the health status of numerically controlled machining tools based on multi-sensor fusion includes a data acquisition module, a noise data preprocessing module, a noise feature analysis and pattern recognition module, a tool health status evaluation module, and a multi-source data fusion module. The data acquisition module, the noise data preprocessing module, the noise feature analysis and pattern recognition module, and the tool health status evaluation module are connected in sequence. Both the data acquisition module and the tool health status evaluation module are connected to the multi-source data fusion module, and the multi-source data fusion module is connected to the numerical control system.

[0007] Preferably, the data acquisition module is used to collect a variety of physical quantities related to the tool health state, and is configured with a vibration sensor, a force sensor, an acoustic emission sensor and a temperature sensor.

[0008] Preferably, the vibration sensor uses a triaxial MEMS accelerometer and is installed at the spindle bearing seat; the force sensor is a dynamic cutting force measurement unit integrated inside the intelligent tool holder; the acoustic emission sensor is fixed to the side of the workbench by magnetic attraction; the temperature sensor is an infrared thermal imager arranged 300 - 350 mm above the processing area.

[0009] Preferably, the noise data preprocessing module preprocesses the collected raw data, removes high-frequency noise, smooths data fluctuations, enhances useful signals, and improves data quality. Specifically, filtering, smoothing and signal enhancement technologies are adopted. For example, a low-pass filter is used to remove high-frequency noise, and the moving average method is used to smooth data fluctuations. At the same time, the data of each sensor is normalized to prepare for subsequent feature extraction and data fusion.

[0010] Preferably, the noise feature analysis and pattern recognition module extracts features closely related to the tool health state from the preprocessed data, such as time-domain features (mean, mean square value, standard deviation, etc.), frequency-domain features (mean of power spectrum, skewness of power spectrum, etc.) and time-frequency domain features (wavelet packet energy band, etc.). The random forest algorithm and the XGBoost algorithm are used to screen the extracted features respectively, and then the union of the features screened by the two is taken for fusion to reduce the model calculation time and obtain features with strong correlation with the tool health state. At the same time, machine learning models such as deep belief network (DBN) and convolutional neural network (CNN) are used to perform pattern recognition on the tool health state.

[0011] Preferably, the tool health state evaluation module evaluates the tool health state according to the results of feature analysis and pattern recognition, combined with the preset tool health state evaluation criteria. The tool health state is divided into different levels such as initial wear, mild wear, moderate wear, severe wear and tool failure. This module pre-divides the tool wear state according to the tool wear value, obtains the corresponding tool wear state according to the output tool wear value, and compares it with the tool wear state output by the machine learning model to determine the finally evaluated tool health state.

[0012] Preferably, the multi-source data fusion module uses the weighted average algorithm to fuse the data collected by multiple sensors. The weights are calculated according to the reliability and accuracy of each sensor, and weighted average fusion is performed to generate high-quality fused data, so as to comprehensively utilize the information of multiple sensors and improve the accuracy of tool health state evaluation.

[0013] Online evaluation method for the health state of CNC machining tools based on multi-sensor fusion, comprising the following steps:

[0014] 1. Data acquisition: During the CNC machining process, use the data acquisition module to collect various physical quantities related to the health state of the tool, such as vibration signals, cutting force signals, acoustic emission signals, temperature signals, etc.

[0015] 2. Data preprocessing: Perform preprocessing on the collected raw data for noise data, and use filtering, smoothing, and signal enhancement techniques to remove high-frequency noise, smooth data fluctuations, and enhance useful signals. At the same time, perform standardization processing on the data of each sensor to make the data comparable.

[0016] 3. Feature extraction and screening: Extract feature parameters in the time domain, frequency domain, and time-frequency domain from the preprocessed data. Use the random forest algorithm and the XGBoost algorithm to respectively rank the importance of the extracted features, select the top n features with the highest feature importance for each, and then take the union of the selected features to finally obtain m features with strong correlation to the health state of the tool.

[0017] 4. Data fusion: Calculate the weights according to the reliability and accuracy of each sensor, and use the weighted average algorithm to fuse the data collected by multiple sensors to generate high-quality fused data.

[0018] 5. Pattern recognition and health state evaluation: Input the fused features into machine learning models such as deep belief network (DBN) and convolutional neural network (CNN) for training and pattern recognition to obtain the wear state of the tool. At the same time, pre-divide the tool wear values into tool wear states, obtain the corresponding tool wear state according to the output tool wear value, and compare it with the tool wear state output by the machine learning model to determine the finally evaluated health state of the tool.

[0019] 6. Result output and warning: Output the evaluation results in real time to display the current health state of the tool. When the health state of the tool reaches the severe wear or tool failure level, the system issues a warning signal to remind the operator to replace the tool in time.

[0020] Advantages of the present invention:

[0021] 1. Improve evaluation accuracy: By comprehensively integrating data from multiple sensors such as noise, vibration, cutting force, acoustic emission, and temperature, and using the weighted average algorithm for data fusion, the noise interference of a single sensor is reduced, and the accuracy of tool health state evaluation is improved.

[0022] 2. Enhance data quality: Use filtering, smoothing, and signal enhancement techniques to remove high-frequency noise, smooth data fluctuations, and enhance useful signals, improving the data quality and providing a reliable data basis for subsequent feature extraction and pattern recognition.

[0023] 3. Achieve intelligent evaluation: It gets rid of the dependence on signal processing technology and diagnostic experience, uses a machine learning model to achieve the adaptive extraction of tool wear characteristics and the pattern recognition of tool health status, and achieves the purpose of intelligent evaluation.

[0024] 4. Timely warning: It can monitor the health status of the tool in real time, and send a warning signal in time when the tool shows severe wear or failure, avoiding the decline of machining quality and machine tool damage caused by tool problems, improving production efficiency and reducing production costs. Description of the Drawings

[0025] The present invention will be described in detail below in conjunction with the drawings and specific embodiments;

[0026] Figure 1 is the system architecture diagram of the present invention;

[0027] Figure 2 is the method flow chart of the present invention. Specific Embodiments

[0028] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0029] Refer to Figure 1-2 , this specific embodiment adopts the following technical solution: an on-line evaluation system for the health status of CNC machining tools based on multi-sensor fusion, including a data acquisition module 1, a noise data preprocessing module 2, a noise feature analysis and pattern recognition module 3, a tool health status evaluation module 4, and a multi-source data fusion module 5. The data acquisition module 1, the noise data preprocessing module 2, the noise feature analysis and pattern recognition module 3, and the tool health status evaluation module 4 are connected in sequence. Both the data acquisition module 1 and the tool health status evaluation module 4 are connected to the multi-source data fusion module 5, and the multi-source data fusion module 5 is connected to the CNC system.

[0030] It should be noted that the data acquisition module 1 is used to collect various physical quantities related to the tool health state, and is configured with a vibration sensor, a force sensor, an acoustic emission sensor and a temperature sensor. The vibration sensor adopts a three-axis MEMS accelerometer (range ±50g, bandwidth 0.1 - 10kHz) and is installed at the spindle bearing seat; the force sensor is a dynamic cutting force measurement unit (sampling rate 5kHz, accuracy ±0.5% FS) and is integrated inside the intelligent tool holder; the acoustic emission sensor (center frequency 150kHz, sensitivity ≥80dB) is fixed to the side of the workbench by magnetic attraction; the temperature sensor is an infrared thermal imager (resolution 640×480, temperature measurement range 0 - 1000°C) and is arranged 300mm above the processing area; these sensors are used to accurately capture various physical changes of the tool during the processing process.

[0031] It should be noted that the noise data preprocessing module 2 preprocesses the collected raw data, removes high-frequency noise, smooths data fluctuations, enhances useful signals, and improves data quality. Specifically, filtering, smoothing and signal enhancement techniques are adopted. For example, a low-pass filter is used to remove high-frequency noise, and the moving average method is used to smooth data fluctuations. At the same time, the data of each sensor is standardized to prepare for subsequent feature extraction and data fusion. Specifically, it includes: a wavelet threshold denoising unit, which performs 5-layer decomposition using the db8 wavelet basis, and the threshold calculation formula is:

[0032]

[0033]

[0034] where λ is the finally determined threshold; σ is the estimated standard deviation of the noise; N is the length of the signal (number of data points)

[0035] cD1 is the high-frequency coefficient (detail coefficient) after the signal is decomposed by one-level wavelet; 0.6745 is a correction factor used to convert the median absolute deviation (MAD) into the estimated standard deviation of Gaussian noise.

[0036] An adaptive Kalman filter unit, the dynamic adjustment range of the process noise covariance matrix Q is 0.01 - 0.1;

[0037] A signal normalization unit, which performs Z-score normalization:

[0038] where z is the standard score, indicating the degree to which the data point x deviates from the mean (in units of standard deviation);

[0039] x is the original data value;

[0040] μ is the mean of the data population (or sample mean);

[0041] σ is the standard deviation of the data population (or the sample standard deviation).

[0042] It should be noted that the noise feature analysis and pattern recognition module 3 extracts features closely related to the tool health state from the preprocessed data, such as time-domain features (mean, mean square value, standard deviation, etc.), frequency-domain features (mean of power spectrum, skewness of power spectrum, etc.), and time-frequency domain features (wavelet packet energy bands, etc.). The random forest algorithm and the XGBoost algorithm are used to screen the extracted features respectively, and then the union of the features screened by the two is taken for fusion to reduce the model calculation time and obtain features with strong correlation with the tool health state. At the same time, machine learning models such as deep belief network (DBN) and convolutional neural network (CNN) are used to perform pattern recognition on the tool health state.

[0043] The feature extraction unit extracts the following feature groups from the preprocessed signal:

[0044] Time-domain feature group: including 12 indicators such as variance, kurtosis, and waveform factor, and the calculation window width is 200 ms;

[0045] Frequency-domain feature group: The skewness of the power spectrum is calculated using FFT (the number of points is 1024):

[0046]

[0047] Among them, Sk represents the skewness of the power spectrum; N represents the number of samples; f i refers to the power value corresponding to the i-th frequency component; μf is the mean of the power values of all frequency components; P(f i ) represents the probability that the power value f of the i-th frequency component appears i

[0048] Time-frequency feature group: 8 energy sub-bands are obtained through 3-layer wavelet packet decomposition;

[0049] The feature selection unit performs parallel screening using random forest (500 trees) and XGBoost (learning rate 0.01), and takes the union of the features with the top 30% importance rankings of the two;

[0050] The deep learning model unit includes: a deep belief network (DBN) with a structure of 500-300-100-50, trained using the contrastive divergence algorithm; a convolutional neural network (CNN) including 2 convolutional layers (kernel size 5×5) and 3 fully connected layers;

[0051] ​It should be noted that the tool health status evaluation module 4 evaluates the tool health status according to the results of feature analysis and pattern recognition, combined with the preset tool health status evaluation criteria. The tool health status is divided into different levels such as initial wear, mild wear, moderate wear, severe wear, and tool failure. This module pre-divides the tool wear state based on the tool wear value, obtains the corresponding tool wear state according to the output tool wear value, and compares it with the tool wear state output by the machine learning model to determine the finally evaluated tool health status. The health status grading criteria are as follows in the table:

[0052]

[0053] Dual-channel verification mechanism:

[0054] Channel 1: Wear amount VB classification based on physical measurement

[0055] Channel 2: State probability distribution output by the DBN model

[0056] When the difference between the two is > 15%, manual review is triggered;

[0057] It should be noted that the multi-source data fusion module 5 uses the weighted average algorithm to fuse the data collected by multiple sensors. The weights are calculated according to the reliability and accuracy of each sensor, and weighted average fusion is performed to generate high-quality fused data, so as to comprehensively utilize the information of multiple sensors and improve the accuracy of tool health status evaluation.

[0058] The multi-source data fusion module executes the adaptive weighted fusion algorithm:

[0059]

[0060] Where: y fusion represents the final data value after fusion at time t; n represents the number of sensors; w i (t): The weight of the i-th sensor at time t, whose value is between 0 and 1 and satisfies reflecting the proportion of the data of this sensor in the fusion result. x i (t) represents the data value collected by the i-th sensor at time t. ri(t) represents the reliability of the i-th sensor at time t; is the variance of the measurement error of the i-th sensor at time t; For all sensors is summed up as the denominator for normalizing the numerator, so that the sum of the weights of all sensors is 1. The reliability coefficient ri(t) is dynamically adjusted according to the recent failure rate of the sensor (0 - 1);

[0061] Fusion data quality indicators: signal-to-noise ratio ≥ 35 dB (average of the original signal is 20 dB); feature consistency error < 5%.

[0062] In this specific embodiment, vibration sensors, force sensors, acoustic emission sensors, temperature sensors, etc. are installed at appropriate positions on the machine tool to ensure that physical quantities related to the tool health state can be accurately collected. For example, the vibration sensor is installed on the machine tool spindle, the force sensor is installed on the tool shank, the acoustic emission sensor is installed on the machine tool bed, and the temperature sensor is installed near the tool. Connect each sensor to the data acquisition module, and the data acquisition module is connected to a computer or a controller to achieve data acquisition and transmission. Use programming languages (such as Python, MATLAB, etc.) to write the control program and data processing program of the system to achieve functions such as data acquisition, preprocessing, feature extraction, data fusion, pattern recognition, and health state evaluation.

[0063] The online evaluation method for the health state of CNC machining tools based on multi-sensor fusion in this specific embodiment mainly includes:

[0064] 1. Data acquisition: During the CNC machining process, use the data acquisition module to collect various physical quantities related to the tool health state, such as vibration signals, cutting force signals, acoustic emission signals, temperature signals, etc.

[0065] 2. Data preprocessing: Perform preprocessing on the collected raw data for noise data. Use filtering, smoothing, and signal enhancement techniques to remove high-frequency noise, smooth data fluctuations, and enhance useful signals. At the same time, perform standardization processing on the data of each sensor to make the data comparable.

[0066] 3. Feature extraction and screening: Extract feature parameters in the time domain, frequency domain, and time-frequency domain from the preprocessed data. Use the random forest algorithm and the XGBoost algorithm to respectively rank the importance of the extracted features, select the top n features with the highest feature importance for each, and then take the union of the selected features to finally obtain m features with strong correlation to the tool health state.

[0067] 4. Data fusion: Calculate the weights according to the reliability and accuracy of each sensor, and use the weighted average algorithm to fuse the data collected by multiple sensors to generate high-quality fused data.

[0068] 5. Pattern recognition and health state evaluation: Input the fused features into machine learning models such as deep belief network (DBN), convolutional neural network (CNN), etc. for training and pattern recognition to obtain the tool wear state. At the same time, pre-divide the tool wear values for the tool wear states, obtain the corresponding tool wear states according to the output tool wear values, and compare them with the tool wear states output by the machine learning model to determine the finally evaluated tool health state.

[0069] 6. Result Output and Warning: The evaluation results are output in real time to display the current health status of the tool. When the tool health status reaches the level of severe wear or tool failure, the system issues a warning signal to remind the operator to replace the tool in time.

[0070] In this specific embodiment, a closed-loop evaluation system for the health status of CNC machining tools is constructed through multi-sensor collaborative monitoring, intelligent data processing, and machine learning decision-making. Its core working principle is as follows: First, physical quantities in the machining process are collected in real time through multi-source sensors such as vibration, force, acoustic emission, and temperature. After preprocessing such as denoising and standardization, time-domain, frequency-domain, and time-frequency domain features are extracted. Subsequently, the random forest and XGBoost algorithms are used for feature optimization, and the multi-sensor information is integrated through an adaptive weighted fusion algorithm (the weight dynamic adjustment formula is shown in Claim 4). Finally, the fused features are input into a hybrid model composed of a deep belief network (DBN) and a convolutional neural network (CNN) to achieve accurate classification of wear states. The system ensures the reliability of the evaluation through a virtual-real comparison verification mechanism (bidirectional verification of physical measurement values and model prediction values), and triggers a hierarchical warning when the health index (HI) is lower than the threshold.

[0071] This specific embodiment integrates four types of physical quantities (vibration / force / acoustic emission / temperature), with a detection resolution of 2μm. Compared with a single-sensor system, the false alarm rate is reduced by 85%. It adopts an "feature optimization + hybrid deep learning" architecture, and the classification accuracy is ≥96.7% under complex working conditions such as interrupted cutting. It supports standard industrial protocols (OPC UA / MODBUS), and the end-to-end delay from data acquisition to warning output is <80ms, meeting the real-time requirements. The application in an automotive parts production line shows that this system reduces the accidental damage of tools by 72%, extends the average life by 40%, and reduces the comprehensive maintenance cost by 31%.

[0072] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An online evaluation system for the health state of CNC machining tools based on multi-sensor fusion, characterized in that It includes a data acquisition module (1), a noise data preprocessing module (2), a noise feature analysis and pattern recognition module (3), a tool health state evaluation module (4), and a multi-source data fusion module (5). The data acquisition module (1), the noise data preprocessing module (2), the noise feature analysis and pattern recognition module (3), and the tool health state evaluation module (4) are connected in sequence. Both the data acquisition module (1) and the tool health state evaluation module (4) are connected to the multi-source data fusion module (5), and the multi-source data fusion module (5) is connected to the numerical control system.

2. The online evaluation system for the health state of a numerically controlled machining tool based on multi-sensor fusion according to claim 1, wherein The described data acquisition module (1) is used to acquire various physical quantities related to the tool health state, and is configured with vibration sensors, force sensors, acoustic emission sensors, and temperature sensors.

3. The online evaluation system for the health state of a numerically controlled machining tool based on multi-sensor fusion according to claim 1, characterized in that, The vibration sensor uses a three-axis MEMS accelerometer and is installed at the spindle bearing housing; the force sensor is a dynamic cutting force measurement unit integrated inside the intelligent tool holder; the acoustic emission sensor is fixed to the side of the workbench by magnetic attraction; the temperature sensor is an infrared thermal imager arranged 300 - 350 mm above the machining area.

4. The online evaluation system for the health state of a numerically controlled machining tool based on multi-sensor fusion according to claim 1, characterized in that, The described noise data preprocessing module (2) preprocesses the collected raw data, removes high-frequency noise, smooths data fluctuations, enhances useful signals, and improves data quality; at the same time, it standardizes the data of each sensor to prepare for subsequent feature extraction and data fusion.

5. The online evaluation system for the health state of a numerically controlled machining tool based on multi-sensor fusion according to claim 1, characterized in that The described noise feature analysis and pattern recognition module (3) extracts features closely related to the tool health state from the preprocessed data; uses the random forest algorithm and the XGBoost algorithm to screen the extracted features respectively, and then takes the union of the features screened by the two for fusion to reduce the model calculation time and obtain features with strong correlation to the tool health state; at the same time, uses a deep belief network and a convolutional neural network model to perform pattern recognition on the tool health state.

6. The online evaluation system for the health state of a numerically controlled machining tool based on multi-sensor fusion according to claim 1, wherein The described tool health state evaluation module (4) evaluates the tool health state according to the results of feature analysis and pattern recognition, combined with the preset tool health state evaluation criteria; the tool health state is divided into initial wear, mild wear, moderate wear, severe wear, and tool failure levels; This module pre-divides the tool wear state according to the tool wear value, obtains the corresponding tool wear state according to the output tool wear value, and compares it with the tool wear state output by the machine learning model to determine the finally evaluated tool health state.

7. The online evaluation system for the health status of CNC machining tools based on multi-sensor fusion according to claim 1, characterized in that, The described multi-source data fusion module (5) uses a weighted average algorithm to fuse the data collected by multiple sensors; calculates the weights according to the reliability and accuracy of each sensor, performs weighted average fusion, and generates high-quality fused data to comprehensively utilize the information of multiple sensors and improve the accuracy of tool health state evaluation.

8. An online evaluation method for the health state of CNC machining tools based on multi-sensor fusion, characterized in that, It includes the following steps: (1), Data acquisition: During the numerical control machining process, use the data acquisition module to acquire various physical quantities related to the tool health state, such as vibration signals, cutting force signals, acoustic emission signals, temperature signals, etc.; (2), Data preprocessing: Perform noise data preprocessing on the collected raw data, and use filtering, smoothing, and signal enhancement techniques to remove high-frequency noise, smooth data fluctuations, and enhance useful signals; at the same time, standardize the data of each sensor to make the data comparable; (3), Feature extraction and screening: Extract feature parameters in the time domain, frequency domain, and time-frequency domain from the preprocessed data; use the random forest algorithm and the XGBoost algorithm to rank the importance of the extracted features respectively, select the top n features with the highest feature importance for each, and then take the union of the selected features to finally obtain m features that are highly correlated with the tool health state; (4), Data fusion: Calculate weights according to the reliability and accuracy of each sensor, and use the weighted average algorithm to fuse the data collected by multiple sensors to generate high-quality fused data; (5), Pattern recognition and health state assessment: Input the fused features into machine learning models such as deep belief networks and convolutional neural networks for training and pattern recognition to obtain the wear state of the tool; at the same time, pre-divide the tool wear state according to the tool wear value, obtain the corresponding tool wear state based on the output tool wear value, and compare it with the tool wear state output by the machine learning model to determine the finally evaluated tool health state; (6), Result output and warning: Output the evaluation result in real time to display the current health state of the tool; When the tool health state reaches the severe wear or tool failure level, the system issues a warning signal to remind the operator to replace the tool in time.

Images