Intelligent cutter fracture and fatigue detection method based on vibration signal analysis

Abstract

The invention discloses a tool fracture and fatigue intelligent detection method based on vibration signal analysis, and the method comprises the following steps: S1, installing a vibration sensor, and collecting the vibration signal of a tool in real time; s2, the collected tool vibration signals are preprocessed, and noise in the signals is removed; s3, performing time-frequency analysis on the preprocessed vibration signals, and extracting time-frequency features in the signals; s4, performing deep feature learning on the extracted time-frequency features to form deep features; s5, the depth features are classified and analyzed, and the health state of the cutter is output; s6, according to the health state optimization feature extraction and prediction result of the cutter, generating learning output; s7, evaluating the health state of the cutter in real time according to the learning output, and pushing alarm information; and S8, according to the alarm information, predicting the service life of the cutter and optimizing a cutter replacement and maintenance strategy. According to the method, short-time Fourier transform and Hough transform are combined, and the extreme learning machine is applied, so that intelligent detection on the fracture and fatigue of the cutter is realized.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to an intelligent detection method for tool fracture and fatigue based on vibration signal analysis. Background Technology

[0002] In modern manufacturing, the health of cutting tools is crucial to machining accuracy, production efficiency, and equipment safety. During long-term use, cutting tools often experience fatigue, wear, and fracture due to the combined effects of cutting forces, temperature changes, and material hardness. Changes in tool condition typically manifest as abnormal vibration signals, which are closely related to the tool's actual working state. Traditionally, tool condition monitoring relies heavily on manual inspection, periodic replacement, or maintenance based on tool usage time. These methods have several drawbacks, failing to promptly and accurately identify fatigue accumulation or crack formation, easily leading to undetected tool damage, thus affecting machining quality, increasing production costs, and even causing equipment failure. Therefore, developing a method capable of real-time monitoring of tool condition and providing early warnings has become essential for improving production efficiency and equipment safety.

[0003] With the development of technology, more and more researchers are beginning to try using time-frequency analysis methods, such as Short-Time Fourier Transform (STFT) and Wavelet Transform. These methods can simultaneously consider the changes of signals in the time and frequency domains, thereby capturing the instantaneous characteristics of the signal and improving the accuracy of vibration signal analysis. However, even time-frequency analysis methods have certain limitations. In addition, time-frequency analysis methods have high computational complexity, and for large amounts of vibration data generated under complex processing conditions, traditional analysis methods often cannot meet the real-time requirements.

[0004] Building upon vibration signal processing, an increasing number of studies are incorporating intelligent algorithms, particularly machine learning and deep learning methods, to enhance the intelligence level of tool condition monitoring. By learning from historical data, machine learning algorithms can identify the complex relationship between vibration signals and tool health. However, a significant drawback of traditional machine learning methods is their reliance on extensive manual feature extraction, resulting in low automation and requiring highly specialized knowledge from operators. With the rise of deep learning, deep neural networks (DNNs), convolutional neural networks (CNNs), and deep belief networks (DBNs) have become research hotspots in vibration signal analysis. These algorithms, through multi-layered learning structures, can automatically extract high-level features from large amounts of vibration signal data without requiring manual design of complex features. This approach significantly improves the automation and accuracy of tool health monitoring, demonstrating a strong advantage, especially when dealing with complex and nonlinear data.

[0005] While deep learning excels in vibration signal analysis, existing technologies generally lack adaptability. In actual machining, factors such as tool type, machining material, and machining parameters often change, leading to variations in the vibration signal characteristics of the tool. Traditional tool monitoring systems are typically designed for a fixed working condition, making it difficult to achieve good detection results under diverse conditions. Although some research has introduced adaptive algorithms to optimize monitoring systems, the application of these algorithms remains limited, and their adaptability and flexibility are still insufficient.

[0006] Therefore, how to provide an intelligent detection method for tool fracture and fatigue based on vibration signal analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose an intelligent detection method for tool fracture and fatigue based on vibration signal analysis. This invention discloses an intelligent detection method for tool fracture and fatigue based on vibration signal analysis, aiming to detect the health status of tools in real time through precise monitoring and analysis of tool vibration signals, especially early fatigue, damage, and potential fracture. Traditional tool condition monitoring methods often rely on manual inspection or periodic tool replacement, but these methods suffer from long response times and poor monitoring accuracy. This invention, by introducing advanced vibration signal processing technology and intelligent algorithms, can effectively improve the real-time performance and accuracy of tool health status monitoring, reducing production downtime and quality problems caused by tool failures.

[0008] A method for intelligent detection of tool fracture and fatigue based on vibration signal analysis according to an embodiment of the present invention includes the following steps:

[0009] S1. Install multiple high-precision vibration sensors at the contact points between the tool and the machine tool to collect vibration signals of the tool in real time during the machining process, ensuring coverage of the vibration state of the tool under different working conditions;

[0010] S2. Preprocess the acquired vibration signal, including removing noise from the signal and using an adaptive filtering algorithm to remove environmental interference and non-tool-related vibration signals to ensure that the processed signal represents the true vibration characteristics of the tool.

[0011] S3. Perform time-frequency analysis on the preprocessed vibration signal, use short-time Fourier transform to extract the time-frequency features in the signal, and apply Hough transform to perform pattern recognition on the periodic abnormal waveforms in the time-frequency graph to form time-frequency extraction features, capturing frequency changes, periodic fluctuations and weak signals related to the tool health status in the signal.

[0012] S4. Based on time-frequency feature extraction, deep belief networks are used to perform deep feature learning to form deep feature extraction and automatically identify potential damage signals of the tool during operation, especially early signs of tool fatigue, cracks or fractures.

[0013] S5. Based on the deep extracted features, the Extreme Learning Machine is used for classification and analysis, and the health status category of the tool is output, including normal, mild fatigue, severe fatigue and fracture, thereby achieving accurate assessment of the tool status.

[0014] S6. Based on the health status category of the tool, combined with wavelet packet transform and adaptive Boosting algorithm, fine-grained signal patterns are extracted from the health status category of the tool to optimize the tool fatigue and fracture prediction results. For different tool types, processing conditions and working condition changes, the learning output most suitable for the current working condition is generated to ensure high-precision monitoring under different working conditions.

[0015] S7. Based on the learning output, evaluate the status of the tool in real time and push alarm information to the operator in real time to promptly remind them of the potential failure risks of the tool and prioritize alarms by failure type.

[0016] S8. By using alarm information and tool health status assessment, predict the remaining service life of the tool, and combine real-time data analysis to optimize tool replacement and maintenance strategies, ensuring that the tool is processed in the best condition, reducing downtime and improving production efficiency.

[0017] Optionally, S1 specifically includes:

[0018] S11. Multiple high-precision vibration sensors are installed at the contact point between the cutting tool and the machine tool. The sensors are selected to sense high-frequency micro-vibrations to ensure coverage of the vibration state of the cutting tool under different working conditions. The vibration sensors can collect the vibration signal of the cutting tool in real time during the machining process.

[0019] S12. The sensor is connected to the data acquisition module, which includes a signal conditioning circuit, an analog-to-digital converter, and a data storage unit. The signal conditioning circuit is used to amplify, filter, and convert the acquired vibration signal so that the analog-to-digital converter (ADC) can accurately convert the signal into a digital signal.

[0020] S13. The vibration signal is converted into a digital signal by an analog-to-digital converter and the collected digital signal is stored in a data storage unit. The signal stored in the data storage unit is stored in a timestamp manner to ensure the temporal continuity of the signal data.

[0021] S14. The data acquisition module transmits digital signals to the signal processing unit in real time, and the signal processing unit performs preliminary analysis on the acquired vibration signals.

[0022] Optionally, S2 specifically includes:

[0023] S21. The acquired vibration signal is subjected to noise removal processing. The noise includes environmental noise, electromagnetic interference, and mechanical system noise. The noise removal processing adopts an adaptive filtering algorithm, which is the Least Mean Square (LMS) algorithm, and its formula is expressed as:

[0024]

[0025] Where y(n) is the output signal, x(ni) is the delayed version of the input signal, and w i (n) represents the weight coefficients of the filter, M represents the order of the filter, and n represents the time index. In the LMS algorithm, the weight coefficients of the filter are dynamically adjusted according to the error between the signal and the desired signal, and the output of the signal is gradually optimized to achieve the effect of noise removal.

[0026] S22. After filtering, a bandpass filter is used to further process the signal, filtering out unwanted low-frequency and high-frequency noise and retaining the important frequency components in the tool vibration signal.

[0027] S23. The denoised signal is standardized by subtracting its mean and dividing by its standard deviation, so that the denoised signal has zero mean and unit standard deviation, thus ensuring the consistency and stability of the signal in subsequent analysis.

[0028] Optionally, S3 specifically includes:

[0029] S31. Perform time-frequency analysis on the preprocessed vibration signal, using short-time Fourier transform to convert the signal from the time domain to the frequency domain in order to extract the time and frequency characteristics of the vibration signal. The mathematical expression of short-time Fourier transform is:

[0030]

[0031] Where X(t,f) is the time-frequency representation at time t and frequency f, x(τ) is the original vibration signal, w(τ-t) is the window function, and e -j2πfτ The short-time Fourier transform is a complex exponential function of the Fourier transform, which extracts the local features of the vibration signal in the time and frequency domains.

[0032] S32. The time-frequency diagram obtained by the short-time Fourier transform is processed, and then the Hough transform is applied to detect the periodic abnormal waveform in the signal. The Hough transform maps the frequency pattern to a new parameter space, and then through linear parameterization, finds feature points that conform to specific rules in the parameter space, thereby effectively identifying the periodic features in the signal and providing a basis for tool condition judgment. It also identifies periodic features, thereby effectively capturing the small signals of tool condition changes.

[0033] S33. Normalize the processed time-frequency features and compare the features of different frequency bands on a uniform scale to ensure that the influence of each frequency band is balanced in subsequent analysis, thereby improving the accuracy and robustness of feature extraction.

[0034] S34. By analyzing the time-frequency characteristics, the main frequency components and mode changes related to the tool health status are extracted to form time-frequency extraction features.

[0035] Optionally, S4 specifically includes:

[0036] S41. Deep feature learning is performed on the time-frequency extracted features using a deep belief network (DBN). The DBN includes multiple autoencoder layers. Each autoencoder automatically extracts a high-level representation of the input features through unsupervised training. The weights and biases of each layer are optimized by the gradient descent algorithm, and finally, deep features are output for accurate identification of tool status.

[0037] S42. In the deep feature learning process, the network captures tiny signals related to tool fatigue and cracks through unsupervised training, and adjusts the activation function of the autoencoder to extract feature information that is crucial to the judgment of tool status, so as to ensure that the tool status can be accurately extracted from complex vibration signals.

[0038] S43. After deep feature learning is completed, supervised learning is used to fine-tune the DBN model. By inputting the labeled dataset into the model, the relationship between deep features and tool health status is optimized. The loss function is:

[0039]

[0040] in, y represents the tool health status predicted by the model. i The true label is N, and the number of training samples is N. By minimizing the loss function, the accuracy of the model's prediction of the tool state is improved.

[0041] S44. The trained DBN model makes predictions based on the features output by deep learning, forming deep extracted features and obtaining the potential damage signal of the tool.

[0042] Optionally, S5 specifically includes:

[0043] S51. Based on deep feature extraction, the Extreme Learning Machine (ELM) model is used for classification and analysis. ELM randomly generates input layer weights and biases, and uses the least squares method to solve for the output layer weights and biases. The classification result is the tool health status category, including normal, mild fatigue, severe fatigue and fracture, etc.

[0044] S52. Train the ELM model using a training dataset that combines real labels and deep extracted features, and output classification results that minimize the difference from the real labels.

[0045] S53. Perform real-time inference on the trained ELM model, use the model parameters obtained during the training process to classify the newly acquired vibration signals in real time, and output the health status category of the tool.

[0046] S54. Continuously train and update the ELM model, and combine it with other monitoring modules to form a multi-level health assessment system to ensure that the tool's usage status is assessed in a timely and accurate manner.

[0047] Optionally, S6 specifically includes:

[0048] S61. Based on the health status category of the tool output by the extreme learning machine, perform wavelet packet transformation to convert it into a time-frequency vibration signal, and obtain fine-grained time-frequency features through multi-scale wavelet packet decomposition.

[0049] S62. Extract key features from the sub-signals after wavelet packet transform to form a fine-grained pattern of high-frequency changes and signals, and further analyze it to capture subtle changes in tool fatigue or fracture.

[0050] S63. Input the extracted fine-grained features into the adaptive Boosting algorithm (AdaBoost) for classification training. AdaBoost trains weak classifiers through multiple iterations and combines the results of multiple weak classifiers into a weighted average to optimize the prediction results.

[0051] S64. Based on the trained AdaBoost algorithm, the vibration signal characteristics of the tool are classified to determine the health status of the tool and generate the learning output most suitable for the current working condition, which reduces the false alarm rate and ensures that the fatigue and fracture prediction results of the tool have high accuracy and stability.

[0052] Optionally, S7 specifically includes:

[0053] S71. Based on the learning output, generate a real-time evaluation result of the current operating status of the tool;

[0054] S72. Calculate the alarm priority score based on the health status level and characteristic fluctuation risk information in the assessment results;

[0055] S73. Alarm levels are classified according to the alarm priority score range. Alarm levels include no alarm, minor alarm, moderate alarm and severe alarm.

[0056] S74. Construct an information package containing evaluation results, priority scores, and alarm levels, and push it to the operator through the human-machine interaction system to alert the current operating risks of the tool.

[0057] S75. Record the information packets generated in each assessment to form a historical alarm information sequence arranged in chronological order for retrospective analysis and trend judgment.

[0058] Optionally, S8 specifically includes:

[0059] S81. Calculate the remaining tool life based on real-time monitoring data and deep learning model output. The calculation method is the ratio of remaining life to maximum tool life, based on the current tool health status score and life change rate.

[0060] S82. Based on the remaining service life prediction results, determine whether the tool needs to be replaced or maintained immediately. If the remaining service life is lower than the preset threshold, trigger the replacement or maintenance operation.

[0061] S83. Calculate the cumulative fatigue level of the tool by summing the ratio of the health status score of each measurement to the maximum lifespan to assess the fatigue damage level of the tool.

[0062] S84. Based on the degree of fatigue accumulation and remaining service life, optimize the tool replacement strategy and recommend the optimal replacement time to reduce downtime in production.

[0063] S85 generates automated tool maintenance and replacement plans, combining remaining service life and fatigue accumulation results, and pushes them to operators to ensure that tools are always in optimal condition.

[0064] S86. Operators adjust maintenance strategies based on real-time data and forecast results, optimize tool management, and ensure that tools do not affect machining quality due to excessive wear or breakage, thereby improving production efficiency and equipment safety.

[0065] The beneficial effects of this invention are:

[0066] This invention presents an intelligent detection method for tool fracture and fatigue based on vibration signal analysis, which effectively improves the accuracy and real-time performance of tool health monitoring and solves a series of problems existing in the prior art. Firstly, by combining short-time Fourier transform and Hough transform time-frequency analysis methods, subtle changes in vibration signals can be accurately captured, especially in the early stages of tool fatigue or damage. Traditional methods often fail to identify these minute signal changes in a timely manner, while this invention, through multi-dimensional time-frequency feature extraction, effectively improves the early warning capability of tool faults, greatly reducing missed or false alarms.

[0067] Secondly, advanced intelligent algorithms such as deep belief networks and extreme learning machines are employed to achieve deep feature learning and automatic classification of tool conditions. By automatically extracting key features from vibration signals and performing efficient classification, this invention significantly improves the accuracy of tool health status assessment. Compared with traditional methods, deep learning and machine learning algorithms can automatically identify minor tool damage, fatigue accumulation, and the risk of impending fracture without human intervention. This not only reduces reliance on manual intervention but also greatly improves the automation and accuracy of fault detection.

[0068] Furthermore, this invention achieves adaptive optimization of the tool monitoring system. Under different machining conditions, the system can automatically adjust algorithm parameters and feature extraction methods to ensure high-precision monitoring under various working conditions. This adaptability solves the problem of reduced detection accuracy caused by changes in working conditions in traditional tool monitoring methods, ensuring the stability and reliability of the system in complex production environments.

[0069] Furthermore, by combining remaining service life prediction and fatigue accumulation analysis, this invention can provide optimized solutions for tool maintenance and replacement. By calculating the tool's health status and predicting its service life in real time, the system can provide early warnings and reasonable maintenance suggestions before tool failure occurs. This intelligent tool maintenance strategy not only reduces downtime in production but also effectively lowers maintenance costs and improves production efficiency and equipment utilization.

[0070] Finally, this invention helps operators promptly identify potential problems through real-time assessment of tool health and an intelligent fault early warning mechanism, avoiding production interruptions and equipment damage caused by tool breakage or severe wear. Combined with data visualization and detailed fault analysis reports, operators can make quick decisions to ensure tools are always in optimal condition, thereby improving machining quality and equipment safety.

[0071] In summary, this invention provides a high-precision, high-real-time, and intelligent tool health monitoring method by combining vibration signal analysis, deep learning and machine learning algorithms, as well as intelligent early warning and maintenance optimization strategies. This method greatly improves the efficiency of tool management, reduces downtime and production losses, and has significant beneficial effects. Attached Figure Description

[0072] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0073] Figure 1 This is a flowchart of an intelligent detection method for tool fracture and fatigue based on vibration signal analysis proposed in this invention;

[0074] Figure 2 This is a schematic diagram of the main algorithm structure of the intelligent detection method for tool fracture and fatigue based on vibration signal analysis proposed in this invention. Detailed Implementation

[0075] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0076] refer to Figure 1-2 A method for intelligent detection of tool fracture and fatigue based on vibration signal analysis includes the following steps:

[0077] S1. Install multiple high-precision vibration sensors at the contact points between the tool and the machine tool to collect vibration signals of the tool in real time during the machining process, ensuring coverage of the vibration state of the tool under different working conditions;

[0078] S2. Preprocess the acquired vibration signal, including removing noise from the signal and using an adaptive filtering algorithm to remove environmental interference and non-tool-related vibration signals to ensure that the processed signal represents the true vibration characteristics of the tool.

[0079] S3. Perform time-frequency analysis on the preprocessed vibration signal, use short-time Fourier transform to extract the time-frequency features in the signal, and apply Hough transform to perform pattern recognition on the periodic abnormal waveforms in the time-frequency graph to form time-frequency extraction features, capturing frequency changes, periodic fluctuations and weak signals related to the tool health status in the signal.

[0080] S4. Based on time-frequency feature extraction, deep belief networks are used to perform deep feature learning to form deep feature extraction and automatically identify potential damage signals of the tool during operation, especially early signs of tool fatigue, cracks or fractures.

[0081] S5. Based on the deep extracted features, the Extreme Learning Machine is used for classification and analysis, and the health status category of the tool is output, including normal, mild fatigue, severe fatigue and fracture, thereby achieving accurate assessment of the tool status.

[0082] S6. Based on the health status category of the tool, combined with wavelet packet transform and adaptive Boosting algorithm, fine-grained signal patterns are extracted from the health status category of the tool to optimize the tool fatigue and fracture prediction results. For different tool types, processing conditions and working condition changes, the learning output most suitable for the current working condition is generated to ensure high-precision monitoring under different working conditions.

[0083] S7. Based on the learning output, evaluate the status of the tool in real time and push alarm information to the operator in real time to promptly remind them of the potential failure risks of the tool and prioritize alarms by failure type.

[0084] S8. By using alarm information and tool health status assessment, predict the remaining service life of the tool, and combine real-time data analysis to optimize tool replacement and maintenance strategies, ensuring that the tool is processed in the best condition, reducing downtime and improving production efficiency.

[0085] In this embodiment, S1 specifically includes:

[0086] S11. Multiple high-precision vibration sensors are installed at the contact point between the cutting tool and the machine tool. The sensors are selected to sense high-frequency micro-vibrations to ensure coverage of the vibration state of the cutting tool under different working conditions. The vibration sensors can collect the vibration signal of the cutting tool in real time during the machining process.

[0087] S12. The sensor is connected to the data acquisition module, which includes a signal conditioning circuit, an analog-to-digital converter, and a data storage unit. The signal conditioning circuit is used to amplify, filter, and convert the acquired vibration signal so that the analog-to-digital converter (ADC) can accurately convert the signal into a digital signal.

[0088] S13. The vibration signal is converted into a digital signal by an analog-to-digital converter and the collected digital signal is stored in a data storage unit. The signal stored in the data storage unit is stored in a timestamp manner to ensure the temporal continuity of the signal data.

[0089] S14. The data acquisition module transmits digital signals to the signal processing unit in real time, and the signal processing unit performs preliminary analysis on the acquired vibration signals.

[0090] This invention provides a tool health intelligent detection method based on vibration signal analysis. By monitoring the tool vibration signal in real time, the method accurately assesses the tool health status and provides timely warnings of potential faults. The method collects vibration signals and performs noise removal and filtering to ensure signal accuracy. Then, it combines short-time Fourier transform and Hough transform for time-frequency analysis to extract the characteristics of tool fatigue and damage. Next, it uses deep learning to automatically identify damage signals and uses an extreme learning machine to classify the health status.

[0091] In this embodiment, S2 specifically includes:

[0092] S21. The acquired vibration signal is subjected to noise removal processing. The noise includes environmental noise, electromagnetic interference, and mechanical system noise. The noise removal processing adopts an adaptive filtering algorithm, which is the Least Mean Square (LMS) algorithm, and its formula is expressed as:

[0093]

[0094] Where y(n) is the output signal, x(ni) is the delayed version of the input signal, and w i (n) represents the weight coefficients of the filter, M represents the order of the filter, and n represents the time index. In the LMS algorithm, the weight coefficients of the filter are dynamically adjusted according to the error between the signal and the desired signal, and the output of the signal is gradually optimized to achieve the effect of noise removal.

[0095] S22. After filtering, a bandpass filter is used to further process the signal, filtering out unwanted low-frequency and high-frequency noise and retaining the important frequency components in the tool vibration signal.

[0096] S23. The denoised signal is standardized by subtracting its mean and dividing by its standard deviation, so that the denoised signal has zero mean and unit standard deviation, thus ensuring the consistency and stability of the signal in subsequent analysis.

[0097] In the vibration signal preprocessing process, this invention first dynamically suppresses environmental noise, electromagnetic interference, and mechanical system noise based on the least mean square (LMS) adaptive filtering algorithm. By continuously optimizing the weighting coefficients according to the error between the signal and the desired signal, the quality of the output signal is effectively improved. Subsequently, a bandpass filter is introduced to further filter out irrelevant low-frequency and high-frequency components in the signal, accurately retaining key frequency band information reflecting the tool status, thereby enhancing the signal discrimination capability from the perspective of spectral structure. Finally, a standardization method with zero mean and unit standard deviation is adopted to unify the signal distribution pattern, improve the stability and compatibility of subsequent feature extraction and model training stages, and provide a high-quality input foundation with strong consistency and high anti-interference capability for intelligent diagnostic models.

[0098] In this embodiment, S3 specifically includes:

[0099] S31. Perform time-frequency analysis on the preprocessed vibration signal, using short-time Fourier transform to convert the signal from the time domain to the frequency domain in order to extract the time and frequency characteristics of the vibration signal. The mathematical expression of short-time Fourier transform is:

[0100]

[0101] Where X(t,f) is the time-frequency representation at time t and frequency f, x(τ) is the original vibration signal, w(τ-t) is the window function, and e -j2πfτ The short-time Fourier transform is a complex exponential function of the Fourier transform, which extracts the local features of the vibration signal in the time and frequency domains.

[0102] S32. The time-frequency diagram obtained by the short-time Fourier transform is processed, and then the Hough transform is applied to detect the periodic abnormal waveform in the signal. The Hough transform maps the frequency pattern to a new parameter space, and then through linear parameterization, finds feature points that conform to specific rules in the parameter space, thereby effectively identifying the periodic features in the signal and providing a basis for tool condition judgment. It also identifies periodic features, thereby effectively capturing the small signals of tool condition changes.

[0103] S33. Normalize the processed time-frequency features and compare the features of different frequency bands on a uniform scale to ensure that the influence of each frequency band is balanced in subsequent analysis, thereby improving the accuracy and robustness of feature extraction.

[0104] S34. By analyzing the time-frequency characteristics, the main frequency components and mode changes related to the tool health status are extracted to form time-frequency extraction features.

[0105] This invention proposes a feature extraction mechanism combining time-frequency analysis and pattern recognition. First, the vibration signal is converted from the time domain to the frequency domain using short-time Fourier transform to extract local features of the signal in time and frequency. Then, Hough transform is applied to detect periodic abnormal waveforms in the time-frequency graph, identify periodic features in the signal, and capture weak signals of tool condition changes. Next, the time-frequency features are normalized to ensure that features in different frequency bands are compared on a uniform scale, improving the accuracy and robustness of feature extraction. Finally, the extracted time-frequency features are analyzed to identify the main frequency components and pattern changes related to the tool health status, providing accurate feature input for subsequent fault diagnosis.

[0106] In this embodiment, S4 specifically includes:

[0107] S41. Deep feature learning is performed on the time-frequency extracted features using a deep belief network (DBN). The DBN includes multiple autoencoder layers. Each autoencoder automatically extracts a high-level representation of the input features through unsupervised training. The weights and biases of each layer are optimized by the gradient descent algorithm, and finally, deep features are output for accurate identification of tool status.

[0108] S42. In the deep feature learning process, the network captures tiny signals related to tool fatigue and cracks through unsupervised training, and adjusts the activation function of the autoencoder to extract feature information that is crucial to the judgment of tool status, so as to ensure that the tool status can be accurately extracted from complex vibration signals.

[0109] S43. After deep feature learning is completed, supervised learning is used to fine-tune the DBN model. By inputting the labeled dataset into the model, the relationship between deep features and tool health status is optimized. The loss function is:

[0110]

[0111] in, y represents the tool health status predicted by the model. i The true label is N, and the number of training samples is N. By minimizing the loss function, the accuracy of the model's prediction of the tool state is improved.

[0112] S44. The trained DBN model makes predictions based on the features output by deep learning, forming deep extracted features and obtaining the potential damage signal of the tool.

[0113] This invention proposes a deep feature learning method based on deep belief networks (DBNs) for tool condition assessment. Through unsupervised training of a multi-layer autoencoder, high-level representations of time-frequency features are automatically extracted, and weights and biases are optimized to achieve accurate tool condition identification. During training, the DBN captures subtle signals related to tool fatigue and cracks, and extracts key feature information by adjusting the activation function. Supervised learning is then used to fine-tune the model, optimizing the relationship between deep features and tool health using labeled data. Minimizing the loss function significantly improves prediction accuracy. The final trained DBN model identifies potential tool damage signals based on deep features, effectively enhancing the model's discriminative ability and robustness, and providing reliable support for accurate tool fault early warning.

[0114] In this embodiment, S5 specifically includes:

[0115] S51. Based on deep feature extraction, the Extreme Learning Machine (ELM) model is used for classification and analysis. ELM randomly generates input layer weights and biases, and uses the least squares method to solve for the output layer weights and biases. The classification result is the tool health status category, including normal, mild fatigue, severe fatigue and fracture, etc.

[0116] S52. Train the ELM model using a training dataset that combines real labels and deep extracted features, and output classification results that minimize the difference from the real labels.

[0117] S53. Perform real-time inference on the trained ELM model, use the model parameters obtained during the training process to classify the newly acquired vibration signals in real time, and output the health status category of the tool.

[0118] S54. Continuously train and update the ELM model, and combine it with other monitoring modules to form a multi-level health assessment system to ensure that the tool's usage status is assessed in a timely and accurate manner.

[0119] This invention proposes a tool health status classification method based on Extreme Learning Machine (ELM) for rapid and efficient identification and evaluation of deeply extracted vibration features. The method classifies tool conditions by randomly initializing input layer weights and biases and using the least squares method to solve for output layer parameters. The output results include multiple categories such as normal, mild fatigue, severe fatigue, and fracture. During model training, a training dataset is constructed by combining real labels and deep features, and classification accuracy is improved by minimizing the difference between the predicted results and the labels. After training, the model can perform real-time inference on newly acquired vibration signals during actual operation, quickly outputting the current tool health status. Furthermore, this method supports continuous model training and dynamic updates, and can operate collaboratively with other monitoring modules to construct a multi-level health assessment system, ensuring the accuracy and timeliness of tool condition identification.

[0120] In this embodiment, S6 specifically includes:

[0121] S61. Based on the health status category of the tool output by the extreme learning machine, perform wavelet packet transformation to convert it into a time-frequency vibration signal, and obtain fine-grained time-frequency features through multi-scale wavelet packet decomposition.

[0122] S62. Extract key features from the sub-signals after wavelet packet transform to form a fine-grained pattern of high-frequency changes and signals, and further analyze it to capture subtle changes in tool fatigue or fracture.

[0123] S63. Input the extracted fine-grained features into the adaptive Boosting algorithm (AdaBoost) for classification training. AdaBoost trains weak classifiers through multiple iterations and combines the results of multiple weak classifiers into a weighted average to optimize the prediction results.

[0124] S64. Based on the trained AdaBoost algorithm, the vibration signal characteristics of the tool are classified to determine the health status of the tool and generate the learning output most suitable for the current working condition, which reduces the false alarm rate and ensures that the fatigue and fracture prediction results of the tool have high accuracy and stability.

[0125] This invention proposes a method for accurately predicting tool health status by combining wavelet packet transform and adaptive Boosting algorithm. First, based on the tool health status category output by the Extreme Learning Machine (ELM), wavelet packet transform is used to convert the vibration signal into a time-frequency signal. Fine-grained time-frequency features are obtained through multi-scale wavelet packet decomposition. Then, key features are extracted from the sub-signals after wavelet packet transform to form high-frequency variation patterns of the signal. Further analysis captures subtle signals of tool fatigue or fracture. The extracted fine-grained features are then input into the adaptive Boosting algorithm for classification training. The adaptive Boosting algorithm iteratively trains a weak classifier and optimizes the prediction results through weighted averaging, classifying the tool vibration signal features, determining the tool health status, and generating the most suitable learning output for the current working condition. This significantly reduces the false alarm rate and ensures high accuracy and stability in tool fatigue and fracture prediction results. This method demonstrates superior accuracy and robustness in tool health monitoring.

[0126] In this embodiment, S7 specifically includes:

[0127] S71. Based on the learning output, generate a real-time evaluation result of the current operating status of the tool;

[0128] S72. Calculate the alarm priority score based on the health status level and characteristic fluctuation risk information in the assessment results;

[0129] S73. Alarm levels are classified according to the alarm priority score range. Alarm levels include no alarm, minor alarm, moderate alarm and severe alarm.

[0130] S74. Construct an information package containing evaluation results, priority scores, and alarm levels, and push it to the operator through the human-machine interaction system to alert the current operating risks of the tool.

[0131] S75. Record the information packets generated in each assessment to form a historical alarm information sequence arranged in chronological order for retrospective analysis and trend judgment.

[0132] This invention proposes a real-time tool operating status assessment and alarm system based on learning output. First, it generates a real-time tool operating status assessment result based on the learning output. Then, based on the health status level and characteristic fluctuation risk information in the assessment result, it calculates an alarm priority score. Next, it classifies alarm levels according to the interval to which the alarm priority score belongs: no alarm, minor alarm, moderate alarm, and severe alarm. Through this process, the system can accurately assess the tool's health status and risk level. Subsequently, it constructs an information package containing the assessment result, priority score, and alarm level, and pushes it to the operator in real time through a human-machine interface system, alerting the operator to the current operational risks of the tool. Finally, the system records the information package generated by each assessment, forming a historical alarm information sequence arranged in chronological order, facilitating subsequent backtracking analysis and trend judgment, thereby helping operators make timely decisions and ensuring the safe and efficient operation of the tool.

[0133] In this embodiment, S8 specifically includes:

[0134] S81. Calculate the remaining tool life based on real-time monitoring data and deep learning model output. The calculation method is the ratio of remaining life to maximum tool life, based on the current tool health status score and life change rate.

[0135] S82. Based on the remaining service life prediction results, determine whether the tool needs to be replaced or maintained immediately. If the remaining service life is lower than the preset threshold, trigger the replacement or maintenance operation.

[0136] S83. Calculate the cumulative fatigue level of the tool by summing the ratio of the health status score of each measurement to the maximum lifespan to assess the fatigue damage level of the tool.

[0137] S84. Based on the degree of fatigue accumulation and remaining service life, optimize the tool replacement strategy and recommend the optimal replacement time to reduce downtime in production.

[0138] S85 generates automated tool maintenance and replacement plans, combining remaining service life and fatigue accumulation results, and pushes them to operators to ensure that tools are always in optimal condition.

[0139] S86. Operators adjust maintenance strategies based on real-time data and forecast results, optimize tool management, and ensure that tools do not affect machining quality due to excessive wear or breakage, thereby improving production efficiency and equipment safety.

[0140] This invention proposes a tool life prediction and maintenance optimization method based on real-time monitoring and deep learning output. First, based on the tool health score and life change rate, the ratio of remaining life to maximum life is calculated to determine the remaining tool life. If the predicted life is lower than a set threshold, replacement or maintenance operations are automatically triggered. Simultaneously, the system assesses tool fatigue by accumulating measured health scores and optimizes replacement strategies based on remaining life, recommending the best replacement time to minimize downtime. Subsequently, the system automatically generates maintenance and replacement plans and pushes the information to operators, ensuring the tool always operates in optimal condition. Finally, operators can flexibly adjust maintenance strategies based on real-time data, achieving intelligent and efficient tool management, improving machining quality and equipment safety.

[0141] Example 1:

[0142] To verify the feasibility of this invention in practice, it was applied to a tool monitoring system in a manufacturing plant in a certain province. This plant focuses on machining high-precision parts, and tool wear and damage directly affect production efficiency and machining quality; therefore, real-time health monitoring and timely maintenance of the tools are of great importance.

[0143] In this factory's production line, the condition of the cutting tools is crucial to the production process. Traditional tool management methods involve periodic inspections and replacement at fixed intervals. However, this method often fails to detect potential fatigue or cracks in the tools in a timely manner, leading to tool damage before replacement, causing production downtime and wasted costs. Therefore, the factory decided to introduce an intelligent tool monitoring system based on vibration signal analysis and deep learning to improve tool utilization efficiency and reduce unnecessary downtime. The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in this invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of this invention.

[0144] The factory's production line uses several CNC machine tools for metal cutting. During prolonged machining, the cutting tools are subjected to high-intensity cutting forces and temperature changes, leading to tool fatigue and wear. Traditional inspection methods rely on manual inspection or periodic tool replacement. However, because tool wear is usually slow and difficult to detect, the optimal replacement time is often missed, resulting in decreased machining quality and even breakage failures. In severe cases, this can cause the entire production line to shut down.

[0145] To address this issue, this invention provides a tool health monitoring method based on vibration signal analysis, deep learning, and real-time fault early warning. By collecting tool vibration signals in real time and using a deep learning model to predict the remaining tool life, the method optimizes tool maintenance and replacement strategies, improves production efficiency, and reduces the risk of failure.

[0146] In this factory, multiple high-precision vibration sensors are installed at the contact points between the cutting tool and the machine tool to collect vibration signals of the tool in real time. The vibration signals first undergo noise removal processing, using an adaptive filtering algorithm to remove environmental noise, electromagnetic interference, and mechanical system noise, ensuring signal accuracy. Next, time-frequency analysis is performed using short-time Fourier transform and Hough transform to extract local features of the vibration signal in time and frequency. Finally, deep feature learning is performed through a deep belief network to automatically identify potential damage to the cutting tool.

[0147] Extreme Learning Machine (ELM) is used to classify extracted deep features and predict tool health status, including normal, mild fatigue, severe fatigue, and fracture. Wavelet packet transform and adaptive Boosting algorithms are combined to further optimize the tool health status prediction results, ensuring high accuracy and low false alarm rate in tool fault warnings. Based on the tool health status prediction results, the system assesses the remaining tool life in real time and generates automated tool maintenance and replacement plans, which are then pushed to operators to ensure the tool is always in optimal working condition.

[0148] To verify the effectiveness of this method, a three-month experiment was conducted in a factory. During the experiment, the usage of 100 cutting tools was monitored, and the traditional periodic inspection method was compared with the intelligent monitoring system of this invention. The data table is shown below:

[0149]

[0150] As can be seen from the data in Table 1, the tool health monitoring system of this invention significantly improves the accuracy of tool failure prediction and effectively reduces production downtime. According to experimental data, the average prediction error for remaining tool life is 6.5 hours, demonstrating the system's ability to accurately predict tool health status. Specifically, the prediction errors for tools numbered 001, 005, and 009 are all within 5 hours, indicating that the system can accurately predict the remaining tool life, ensuring that tools are replaced at the optimal time. For tools numbered 003, 007, 002, and 006, the prediction errors are mostly controlled between 5 and 20 hours, allowing for early detection of tool fatigue signs and further optimizing the maintenance plan.

[0151] During the three-month trial period, a total of 100 cutting tools were monitored. The system successfully predicted 95% of the faulty tools and kept the false alarm rate below 5%, far lower than that of traditional methods. Furthermore, the system's intelligent prediction helped the factory reduce production downtime by 25%, significantly improving the overall efficiency of the production line. This demonstrates that the tool health monitoring system of this invention not only improves the accuracy of tool health status assessment but also effectively reduces production downtime, providing factories with a more efficient production management solution.

[0152] In summary, the tool health monitoring system provided by this invention has demonstrated significant beneficial effects in practical applications, including improving the accuracy of fault prediction, reducing downtime, optimizing maintenance strategies, and providing a more efficient solution for tool management.

[0153] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent detection of tool fracture and fatigue based on vibration signal analysis, characterized in that, Includes the following steps: S1. Install vibration sensors at the contact points between the cutting tool and the machine tool to collect vibration signals of the cutting tool during the machining process in real time; S2. Preprocess the acquired tool vibration signal and use an adaptive filtering algorithm to remove environmental interference and non-tool-related vibration signals; S3. Perform time-frequency analysis on the preprocessed vibration signal, use short-time Fourier transform to extract the time-frequency features in the signal, and apply Hough transform to perform pattern recognition on the periodic abnormal waveforms in the time-frequency graph to form time-frequency extracted features. S4. Based on time-frequency feature extraction, deep belief networks are used to perform deep feature learning, forming deep feature extraction and automatically identifying potential damage signals of the tool during operation. S5. Based on the deep extracted features, the Extreme Learning Machine model is used for classification and analysis, and the health status category of the tool is output. S6. Based on the health status category of the tool, combined with wavelet packet transform and adaptive Boosting algorithm, extract fine-grained signal patterns from the health status category of the tool, optimize the tool fatigue and fracture prediction results, and generate the learning output most suitable for the current working condition. S7. Based on the learning output, evaluate the status of the tool in real time and push alarm information to the operator in real time; S8. By using alarm information and tool health status assessment, predict the remaining service life of the tool, and combine real-time data analysis to optimize tool replacement and maintenance strategies.

2. The intelligent detection method for tool fracture and fatigue based on vibration signal analysis according to claim 1, characterized in that, S1 specifically includes: S11. Install a vibration sensor at the contact point between the cutting tool and the machine tool, wherein the sensor is selected as a sensor that senses high-frequency minute vibrations; S12. The sensor is connected to the data acquisition module, which includes a signal conditioning circuit, an analog-to-digital converter, and a data storage unit. S13. The vibration signal is converted into a digital signal by an analog-to-digital converter, and the collected digital signal is stored in a data storage unit. The signal stored in the data storage unit is stored in a timestamp format. S14. The data acquisition module transmits digital signals to the signal processing unit in real time, and the signal processing unit performs preliminary analysis on the acquired vibration signals.

3. The intelligent detection method for tool fracture and fatigue based on vibration signal analysis according to claim 1, characterized in that, S2 specifically includes: S21. The acquired vibration signal is subjected to noise removal using an adaptive filtering algorithm. The adaptive filtering algorithm is the least mean square algorithm, and the formula of the least mean square algorithm is expressed as follows: Where y(n) is the output signal, x(ni) is the delayed version of the input signal, and w i (n) represents the weight coefficients of the filter, M represents the order of the filter, and n represents the time index. In the least mean square algorithm, the weight coefficients of the filter are dynamically adjusted according to the error between the signal and the desired signal. S22. After adaptive filtering, a bandpass filter is used to further process the signal, filtering out unwanted low-frequency and high-frequency noise and retaining the important frequency components in the tool vibration signal. S23. The denoised signal is standardized by subtracting the mean and dividing by the standard deviation, so that the denoised signal has zero mean and unit standard deviation.

4. The intelligent detection method for tool fracture and fatigue based on vibration signal analysis according to claim 1, characterized in that, S3 specifically includes: S31. Perform time-frequency analysis on the preprocessed vibration signal, using short-time Fourier transform to convert the signal from the time domain to the frequency domain, and extract the time and frequency characteristics of the vibration signal. The mathematical expression of short-time Fourier transform is: Where X(t,f) is the time-frequency representation at time t and frequency f, x(τ) is the original vibration signal, w(τ-t) is the window function, and e -j2πfτ The short-time Fourier transform is a complex exponential function of the Fourier transform, which extracts the local features of the vibration signal in the time and frequency domains. S32. Process the time-frequency graph obtained by the short-time Fourier transform, and then apply the Hough transform to detect periodic abnormal waveforms in the signal. The Hough transform maps the frequency pattern to a new parameter space, and then uses linear parameterization to find feature points that conform to specific rules in the parameter space, and effectively identifies periodic features in the signal. S33. Normalize the processed time-frequency features and compare the features of different frequency bands on a unified scale. S34. By analyzing the time-frequency characteristics, the main frequency components and mode changes related to the tool health status are extracted to form time-frequency extraction features.

5. The intelligent detection method for tool fracture and fatigue based on vibration signal analysis according to claim 1, characterized in that, S4 specifically includes: S41. Deep feature learning is performed on time-frequency extracted features using a deep belief network. The deep belief network includes multiple autoencoder layers. Each autoencoder layer automatically extracts a high-level representation of the input features through unsupervised training. The weights and biases of each layer are optimized by the gradient descent algorithm, and finally, deep features are output. S42. In the deep feature learning process, the network captures tiny signals related to tool fatigue and cracks through unsupervised training, and adjusts the activation function of the autoencoder to extract feature information that is crucial for judging the tool status. S43. After deep feature learning is completed, supervised learning is used to fine-tune the deep belief network model. By inputting the labeled dataset into the model, the relationship between deep features and tool health status is optimized. The loss function is: in, y represents the tool health status predicted by the model. i The true labels are N, and the number of training samples is N. S44. The trained deep belief network model makes predictions based on the features output by deep learning, forming deep extracted features and obtaining the potential damage signal of the tool.

6. The intelligent detection method for tool fracture and fatigue based on vibration signal analysis according to claim 1, characterized in that, S5 specifically includes: S51. Based on deep feature extraction, the Extreme Learning Machine (ELM) model is used for classification and analysis. The ELM randomly generates input layer weights and biases, and uses the least squares method to solve for the output layer weights and biases. The classification result is the health status category of the tool. S52. The Extreme Learning Machine model is trained using a training dataset that combines real labels and deep feature extraction, and the output classification result is designed to minimize the difference between the real labels and the actual labels. S53. Perform real-time inference on the trained extreme learning machine model, use the model parameters obtained during the training process to classify the newly acquired vibration signals in real time, and output the health status category of the tool. S54. Continuously train and update the Extreme Learning Machine model to form a multi-level health assessment system.

7. The intelligent detection method for tool fracture and fatigue based on vibration signal analysis according to claim 1, characterized in that, S6 specifically includes: S61. Based on the health status category of the tool output by the extreme learning machine, perform wavelet packet transformation to convert it into a time-frequency vibration signal, and obtain fine-grained time-frequency features through multi-scale wavelet packet decomposition. S62. Extract key features from the sub-signals after wavelet packet transform to form fine-grained patterns of high-frequency changes and signals; S63. Input the extracted fine-grained features into the adaptive Boosting algorithm for classification training. The adaptive Boosting algorithm trains weak classifiers through multiple iterations and combines the results of multiple weak classifiers into a weighted average to optimize the prediction results. S64. Based on the trained adaptive Boosting algorithm, classify the vibration signal characteristics of the tool to further accurately determine the health status of the tool and generate the learning output most suitable for the current working condition.

8. The intelligent detection method for tool fracture and fatigue based on vibration signal analysis according to claim 1, characterized in that, Specifically, S7 includes: S71. Based on the learning output, generate a real-time evaluation result of the current operating status of the tool; S72. Calculate the alarm priority score based on the health status level and characteristic fluctuation risk information in the assessment results; S73. Alarm levels are classified according to the alarm priority score range. Alarm levels include no alarm, minor alarm, moderate alarm and severe alarm. S74. Construct an information package containing evaluation results, priority scores, and alarm levels, and push it to the operator through the human-machine interaction system to alert the tool's current operational risks. S75. Record the information packets generated in each assessment to form a historical alarm information sequence arranged in chronological order.

9. The intelligent detection method for tool fracture and fatigue based on vibration signal analysis according to claim 1, characterized in that, S8 specifically includes: S81. Calculate the remaining tool life based on real-time monitoring data and deep learning model output. The calculation method is the ratio of remaining life to maximum tool life, based on the current tool health status score and life change rate. S82. Based on the predicted remaining service life, determine the tool's lifespan. If the remaining tool lifespan is lower than a preset threshold, trigger a replacement and maintenance operation. S83. Calculate the cumulative fatigue level of the tool by summing the ratio of the health status score of each measurement to the maximum lifespan to assess the fatigue damage level of the tool. S84. Based on the degree of fatigue accumulation and remaining service life, optimize the tool replacement strategy and recommend the optimal replacement time; S85 generates automated tool maintenance and replacement plans, combining remaining tool life and fatigue accumulation results, and pushes them to operators; S86. Operators adjust maintenance strategies and optimize tool management based on real-time data and forecast results.

Images