Machine tool processing tool abnormal sound monitoring and shutdown control method

Abstract

This invention provides a method for monitoring and controlling abnormal noise from machine tool processing tools, relating to the field of intelligent monitoring technology in mechanical processing. The method includes the following steps: S1: Pre-collecting and storing various typical fault audio samples; S2: Real-time sound acquisition; S3: Sound signal preprocessing, performing noise reduction on the acquired raw sound signal; S4: Real-time spectrum analysis, performing a fast Fourier transform on the preprocessed sound signal to extract the spectral feature vector; S5: Dual anomaly detection, employing a dual judgment logic of real-time comparison and feature matching for anomaly detection; S6: Stop control triggering, when the anomaly judgment condition is met, the central processing unit issues a stop command to the machine tool control system. This invention adopts a high-speed sound acquisition and parallel processing architecture, with a response time of ≤50ms from the occurrence of abnormal sound to the issuance of the stop command. It can achieve stop protection at the moment of tool collision or breakage, avoiding continuous impact damage to the machine tool spindle.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology in machining, and in particular to a method for monitoring abnormal noise from machine tool cutting tools and controlling machine shutdown. Background Technology

[0002] With the rapid development of intelligent manufacturing technology, CNC machine tools have become the core equipment of modern machinery manufacturing. In actual machining processes, the cutting tool, as a key component directly involved in cutting, directly affects machining quality, production efficiency, and equipment safety. Under conditions such as high-speed cutting, heavy-duty machining, and unattended operation, cutting tools frequently experience sudden malfunctions such as collisions, breakage, and abnormal wear. If these malfunctions are not detected and stopped in time, they will cause serious consequences.

[0003] Current protection mechanisms for CNC machine tools mainly rely on traditional methods such as overload protection and travel limit switches. These protection methods have significant time lags: when a tool collides or breaks, irreversible damage has already been caused to core components such as the machine tool spindle and guideways. Traditional tool life management is based solely on machining time or the number of cuts for prediction, which cannot cope with sudden tool breakage or collision accidents. Secondly, existing acoustic emission monitoring systems generally suffer from problems such as difficulty in setting thresholds, high false alarm rates, and response delays, making it difficult to meet the high-precision monitoring needs of actual industrial sites. In addition, they cannot achieve truly unattended machining; operators must monitor the machining status in real time, resulting in high labor costs and low efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a method for monitoring abnormal noise from machine tool processing tools and controlling machine shutdown, thereby solving the technical problems existing in the prior art.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] A method for monitoring abnormal noise from machine tool processing tools and controlling machine shutdown includes the following steps: S1: Constructing an audio sample library, pre-collecting and storing various typical fault audio samples; S2: Real-time sound acquisition, continuously and at high speed acquiring sound signals from the machine tool processing area using a high-sensitivity sound acquisition device, with a sampling frequency of not less than 44.1kHz and a sampling accuracy of not less than 16 bits; S3: Sound signal preprocessing, performing noise reduction processing on the acquired raw sound signals to remove machine tool background noise and environmental interference, extracting effective cutting sound signals, and using wavelet transform algorithm for signal denoising; S4: Real-time spectrum analysis, performing fast Fourier transform on the preprocessed sound signals to convert the time-domain signal into a frequency-domain signal, and extracting the spectral feature vector of the sound signal, including fundamental frequency, harmonic distribution, energy distribution, and peak characteristic parameters; S5: Dual anomaly judgment, using a dual judgment logic of real-time comparison and feature matching for anomaly detection; S6: Shutdown control triggering, when the anomaly judgment conditions are met, the central processing unit sends a shutdown command to the machine tool control system.

[0007] Furthermore, it also includes S7: audio data management, which automatically records and stores sound data before and after an anomaly occurs, supporting post-event playback and analysis.

[0008] Furthermore, step S1 includes tool collision sound, tool breakage sound, abnormal tool wear sound, and abnormal spindle vibration sound.

[0009] Furthermore, the noise reduction process in step S3 employs a wavelet transform noise reduction algorithm, the formula of which is:

[0010] In the formula, The original sound signal. For wavelet transform operators, For frequency domain masking functions, This is a frequency domain multiplication operation. For wavelet inverse transform operators, This is the effective audio signal after noise reduction.

[0011] Furthermore, the spectral analysis of the Fast Fourier Transform in step S4 uses the following formula:

[0012] In the formula, It is a time-domain sound signal. Angular frequency, The imaginary unit, It is a frequency domain spectral function; the spectral energy entropy value is calculated. To determine the degree of abnormality in the sound signal.

[0013] Furthermore, the dual anomaly judgment mechanism in step S5 specifically includes: S501, real-time comparison judgment: the amplitude, frequency, and energy parameters of the current sound signal are compared with the reference threshold during normal processing in real time. When the sound amplitude suddenly increases by more than 60dB above the preset threshold or the frequency characteristics change abruptly, it is judged as a level one anomaly; S502, feature matching judgment: the spectral feature vector of the current sound signal is matched with the feature template in the fault audio sample library for similarity calculation. The cosine similarity algorithm is used. When the similarity is ≥0.85, it is judged as a level two anomaly.

[0014] Furthermore, the shutdown control triggering condition in step S6 is: the duration of a first-level abnormality exceeds 10ms or a second-level abnormality is successfully matched; the shutdown command adopts a dual redundancy mechanism of hard-wired signals and communication commands.

[0015] Furthermore, in step S7, sound data is automatically recorded and stored for 5 seconds before and after the anomaly occurs, while also supporting the user to freely turn the monitoring function on and off and flexibly configure various parameter thresholds.

[0016] Furthermore, step S2 employs: a MEMS microphone array installed in the machine tool machining bay near the spindle; a central processing unit using an ARM Cortex-M7 architecture microprocessor running a real-time operating system; and a machine tool control interface supporting multiple communication methods including IO hardwiring, Modbus bus, and OPC UA.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] (i) The present invention adopts a high-speed sound acquisition and parallel processing architecture. The response time from the occurrence of abnormal sound to the issuance of a stop command is ≤50ms. It can realize stop protection at the moment of tool collision or breakage, effectively avoiding continuous impact damage to the machine tool spindle.

[0019] (ii) This invention uses a dual anomaly judgment mechanism of real-time threshold comparison and spectral feature matching, combined with wavelet transform noise reduction technology, to effectively distinguish between normal cutting noise and fault noise, thus ensuring safety and avoiding unnecessary production interruptions.

[0020] (III) The present invention has a built-in standardized fault audio sample library, covering typical fault scenarios such as collision, fracture, and abnormal wear of various cutting tools such as milling cutters, drills, fly cutters, and taps. It is applicable to various CNC machine tools such as machining centers, CNC lathes, and gantry machining centers, and has strong versatility.

[0021] (iv) This invention realizes intelligent and automated monitoring of the processing process, eliminating the need for real-time monitoring by operators, providing key technical support for realizing lights-out factories and unmanned intelligent manufacturing, and improving production efficiency. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for monitoring abnormal noise from machine tool processing tools and controlling machine shutdown, as disclosed in this invention. Detailed Implementation

[0023] To make the content of this invention easier to understand, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Identical components are represented by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "up," and "down" used in the following description refer to directions in the accompanying drawings, while the terms "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0024] Example 1:

[0025] like Figure 1 As shown, this embodiment provides a method for monitoring abnormal noise from machine tool processing tools and controlling machine shutdown, including the following steps:

[0026] S1: Audio sample library construction, pre-collecting and storing various typical fault audio samples; the fault audio samples include tool collision sound, tool breakage sound, abnormal tool wear sound, and abnormal spindle vibration sound; specifically including the following steps:

[0027] S101: The sample library architecture is constructed according to a four-level classification standard. The first level is divided into four categories according to the fault type: tool collision sound, tool breakage sound, abnormal tool wear sound, and abnormal spindle vibration sound. The second level is further subdivided according to the tool type: end mill, ball end mill, drill, fly cutter, tap, turning tool, parting tool, and threading tool. The third level is covered by machine tool type: machining center, CNC lathe, gantry machining center, Swiss-type lathe, and multi-axis machining center. The fourth level distinguishes fault variants under different cutting parameters, materials, and speeds according to operating parameters.

[0028] S102: In a controlled experimental environment, a high-sensitivity microphone of the same model as that deployed on site is used, with a uniform sampling frequency of 48kHz and a sampling accuracy of 24 bits. For each type of fault scenario, at least 10 sets of valid samples are collected under 5 different combinations of cutting parameters, with each set of samples lasting 1-3 seconds, to ensure the representativeness and integrity of the samples.

[0029] S103: Perform a unified preprocessing pipeline on each original audio sample, including wavelet transform noise reduction, endpoint detection, effective segment truncation, and then extract a 16-dimensional spectral feature vector, covering key acoustic features such as fundamental frequency, harmonic coefficients, energy distribution entropy, peak factor, kurtosis, margin factor, spectral centroid, and spectral bandwidth. All feature parameters are standardized according to Z-score to eliminate the influence of dimensions.

[0030] S104: Using four data augmentation techniques—time stretching, pitch transformation, additive noise mixing, and amplitude modulation—the original samples are expanded, generating at least eight variants for each type of original sample, increasing the sample library size by nine times and significantly improving the model's robustness to different environmental noise and different acquisition distances.

[0031] S105: Establish an incremental learning framework for the sample library to support the automatic entry of real fault samples collected by users on-site into the library. The system regularly performs clustering optimization on feature templates, removes redundant samples, and updates classification thresholds to ensure that the sample library continuously adapts to the actual working conditions of users on-site.

[0032] S2: Real-time sound acquisition, continuously and at high speed acquiring sound signals from the machine tool machining area using a high-sensitivity sound acquisition device, with a sampling frequency of no less than 44.1kHz and a sampling accuracy of no less than 16 bits; the sound acquisition device adopts: a MEMS microphone array, each microphone unit with a sensitivity of -38dBFS, a signal-to-noise ratio of ≥65dBA, and a frequency response range of 20Hz-20kHz, the array is arranged in a diamond shape with an element spacing of 15mm, forming spatial directional reception, effectively suppressing ambient noise from the sides and back, and is installed in the machine tool machining chamber near the spindle to ensure that the sound source at the cutting point reaches the sound path without obstruction; the central processing unit adopts an ARM Cortex-M7 architecture microprocessor, runs a real-time operating system, and has a task scheduling delay of ≤1ms, providing hardware acceleration support for signal processing algorithms; the machine tool control interface supports multiple communication methods such as IO hardwiring, Modbus bus, and OPC UA to realize bidirectional data interaction and status feedback with the machine tool CNC system.

[0033] S3: Audio signal preprocessing. This involves noise reduction of the acquired raw audio signal to remove machine tool background noise and environmental interference, extracting the effective cutting sound signal, and using a wavelet transform algorithm for signal denoising. Specifically, this process is applied to the raw signal acquired by the 4-channel microphone. , , , Time delay estimation and beamforming are performed. A delay summation algorithm is used to enhance the acoustic source signal in the cutting point direction and suppress interference in non-target directions, resulting in an enhanced single-channel signal. Using Db4 wavelet basis functions to... A 5-level wavelet packet decomposition was performed to obtain wavelet coefficients at each scale; an improved soft thresholding function was used to threshold the high-frequency coefficients. An adaptive calculation using unbiased likelihood estimation effectively separates the cutting sound signal from the background noise; the noise reduction process employs a wavelet transform noise reduction algorithm, the formula of which is: In the formula, The original sound signal. For wavelet transform operators, For frequency domain masking functions, This is a frequency domain multiplication operation. For wavelet inverse transform operators, The effective sound signal after noise reduction is obtained. Then, a dual-threshold endpoint detection algorithm is used, which combines short-time energy and short-time zero-crossing rate as dual criteria, to accurately locate the start and end points of the cutting sound, extract the effective cutting sound segments, and remove invalid data during the idle period.

[0034] S4: Real-time spectrum analysis. A Fast Fourier Transform (FFT) is performed on the pre-processed audio signal to convert the time-domain signal to the frequency-domain signal, extracting the spectral feature vector of the audio signal, including fundamental frequency, harmonic distribution, energy distribution, and peak characteristic parameters. A 256-point Hamming window sliding window with a step size of 128 points is used to analyze the pre-processed signal. Performing a Fast Fourier Transform (FFT) achieves a continuous conversion from the time domain to the frequency domain. The spectral analysis of the FFT is performed using the following formula: In the formula, It is a time-domain sound signal. Angular frequency, The imaginary unit, It is a frequency domain spectral function;

[0035] Extract a complete set of acoustic features from the spectrum, including time-domain features: short-time energy, short-time average amplitude, zero-crossing rate, peak factor, kurtosis, margin factor, and impulse factor; frequency-domain features: fundamental frequency, harmonic distortion rate, spectral centroid, spectral bandwidth, and spectral flux; and statistical features: spectral energy entropy and the first three dimensions of Mel frequency cepstral coefficients (MFCC).

[0036] By calculating the spectral energy entropy value To determine the degree of abnormality in the sound signal, the formula is as follows: For the first Energy percentage of each frequency sub-band, entropy value The larger the value, the more disordered the spectral distribution and the higher the probability of anomalies;

[0037] It should be noted that during this process, Z-score standardization can be performed on the original 16-dimensional features, and principal component analysis (PCA) can be used to reduce the dimensionality to an 8-dimensional core feature vector, thereby reducing computational complexity and improving real-time performance.

[0038] S5: Dual anomaly detection employs a dual logic of real-time comparison and feature matching for anomaly detection; the dual anomaly detection mechanism specifically includes:

[0039] S501, Real-time comparison and judgment: The amplitude, frequency, and energy parameters of the current sound signal are compared with the reference threshold during normal processing in real time. When the sound amplitude suddenly increases by more than 60dB above the preset threshold or the frequency characteristics change abruptly, it is judged as a level one abnormality.

[0040] 1. The sound amplitude suddenly increases: and ;

[0041] 2. Frequency characteristic abrupt change: ;

[0042] 3. Energy mutation: ;

[0043] S502, Feature matching judgment: The spectral feature vector of the current sound signal is matched with the feature template in the fault audio sample library to calculate the similarity. The cosine similarity algorithm is used. When the similarity is ≥0.85, it is judged as a level 2 anomaly.

[0044] S503 introduces a time smoothing mechanism, which performs a weighted average of the similarity results of three consecutive frames, with confidence weights of 0.2, 0.3, and 0.5, respectively, to avoid misjudgment in a single frame.

[0045] S6: Stop control trigger. When the abnormal judgment conditions are met, the central processing unit sends a stop command to the machine tool control system. Time-debouncing verification is used for first-level abnormalities. When a first-level abnormal state lasts for more than 10ms without any decline, the authenticity of the abnormality is confirmed, and transient interference is eliminated. The stop control trigger conditions are: the duration of a first-level abnormality exceeds 10ms or a second-level abnormality is successfully matched. The stop command adopts a dual redundancy mechanism of hard-wired signals and communication commands. The dual-redundancy stop command adopts a hardware and software dual-redundancy output mechanism, including hard-wired signals and communication commands. The hard-wired signal is an output of an optocoupler-isolated relay, directly connected to the machine tool emergency stop circuit, with a response time ≤1ms. The communication command simultaneously sends a stop code through both Modbus-RTU and OPC UA channels. During this process, graded actions are performed according to the severity of the abnormality: minor abnormalities trigger a feed rate reduction to 25%; moderate abnormalities trigger feed hold; severe abnormalities trigger spindle stop and emergency stop.

[0046] S7: Audio data management, automatically records and stores sound data before and after an anomaly, supporting post-event playback and analysis; the automatic recording and storage of sound data for 5 seconds before and after an anomaly occurs, while also supporting user-controlled on / off monitoring functions and flexible configuration of various parameter thresholds; this invention uses a circular buffer to continuously cache the most recent 10 seconds of raw audio data, with a sampling rate of 48kHz and a buffer capacity ≥1MB; when an anomaly is triggered, the complete audio segment of 10 seconds (5 seconds before and 5 seconds after the anomaly) is automatically transferred and saved to Flash memory with timestamps, each record being approximately 576KB, supporting the storage of ≥1000 anomaly records; each anomaly record is accompanied by metadata, including anomaly type, similarity value, trigger time, machine tool number, tool number, and processing parameters, supporting conditional retrieval and export.

[0047] Example 2:

[0048] This embodiment is implemented on a vertical machining center, and the specific implementation steps are as follows:

[0049] T1: Install the MEMS digital microphone array inside the spindle guard, at the optimal acoustic position of about 30cm from the tool cutting point, with the installation angle aligned with the cutting area to form a 120° conical sound field coverage; collect 30s of sound samples under no-load and normal cutting conditions respectively, establish a reference threshold matrix, set the reference amplitude of normal cutting sound to 40-55dB, and set the abnormal trigger threshold to 65dB;

[0050] T2: Import standardized fault audio sample library, the first level of fault type layer includes 10mm end mill breaking sound The system includes 12 typical fault types such as 8mm drill bit collision noise, severe tool wear noise, and abnormal spindle vibration; the second-level tool type layer covers 6 commonly used tools including end mills, drills, and turning tools; the third-level machine tool type layer adapts to the acoustic characteristics of vertical machining centers; the fourth-level operating condition parameter layer contains fault variants under different speeds and feed rates; each fault type includes 5 feature templates under different operating conditions, totaling 60 standardized feature vectors.

[0051] T3: The sound acquisition unit continuously acquires cutting sound signals, and removes background noise such as machine tool hydraulic pump and coolant pump through wavelet transform algorithm to extract effective cutting sound signals;

[0052] T4: The central processing unit performs a fast Fourier transform every 10ms to calculate the spectral feature vector of the current sound signal, including 16-dimensional feature parameters such as energy distribution, peak frequency, and harmonic coefficients in the range of 0-10kHz.

[0053] T5: When When a 10mm end mill breaks due to excessive cutting allowance, it instantly generates an impact noise with an amplitude of approximately 85dB, and at the same time, the spectral characteristics change drastically. The system first detects a level one anomaly through real-time comparison, and then completes feature matching with the fracture sound sample within 3ms, with a similarity of 0.92, which meets the level two anomaly conditions. The central processing unit immediately sends feed hold and spindle stop commands to the CNC system through hardwiring. The entire response process takes 42ms.

[0054] The results showed that the machine tool stopped immediately after the tool broke, the spindle was not subjected to secondary impact, the broken part of the tool was intact, and the workpiece only had minor damage at the fracture point. After simple repair, it could continue to be used, thus avoiding the scrapping of the workpiece and damage to the spindle caused by continuing to process after the tool broke.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring abnormal noise from machine tool processing tools and controlling machine shutdown, characterized in that: Includes the following steps: S1: Audio sample library construction, pre-collecting and storing various typical fault audio samples; S2: Real-time sound acquisition, continuously and at high speed acquiring sound signals from the machine tool processing area through a high-sensitivity sound acquisition device, with a sampling frequency of not less than 44.1kHz and a sampling accuracy of not less than 16 bits; S3: Sound signal preprocessing, noise reduction processing is performed on the acquired raw sound signal to remove machine tool background noise and environmental interference, extract the effective cutting sound signal, and the wavelet transform algorithm is used for signal denoising; S4: Real-time spectrum analysis, performs fast Fourier transform on the preprocessed audio signal, converts the time domain signal into a frequency domain signal, and extracts the spectral feature vector of the audio signal, including fundamental frequency, harmonic distribution, energy distribution, and peak feature parameters; S5: Dual anomaly detection, employing a dual logic of real-time comparison and feature matching for anomaly detection; S6: Stop control trigger. When the abnormal judgment conditions are met, the central processing unit sends a stop command to the machine tool control system.

2. The method for monitoring and controlling abnormal noise from machine tool processing tools according to claim 1, characterized in that: It also includes S7: Audio Data Management, which automatically records and stores sound data before and after an anomaly occurs, supporting playback and analysis afterwards.

3. The method for monitoring and controlling abnormal noise from machine tool processing tools according to claim 1, characterized in that: Step S1 includes the following sounds: tool collision sound, tool breakage sound, abnormal tool wear sound, and abnormal spindle vibration sound.

4. The method for monitoring and controlling abnormal noise from machine tool processing tools according to claim 3, characterized in that: The noise reduction process in step S3 employs a wavelet transform noise reduction algorithm, the formula of which is: In the formula, The original sound signal. For wavelet transform operators, For frequency domain masking functions, This is a frequency domain multiplication operation. For wavelet inverse transform operators, This is the effective audio signal after noise reduction.

5. The method for monitoring and controlling abnormal noise from machine tool processing tools according to claim 1, characterized in that: The spectral analysis of the Fast Fourier Transform in step S4 uses the following formula: In the formula, It is a time-domain sound signal. Angular frequency, The imaginary unit, It is a frequency domain spectral function; the spectral energy entropy value is calculated. To determine the degree of abnormality in the sound signal.

6. The method for monitoring and controlling abnormal noise from machine tool processing tools according to claim 1, characterized in that: The dual anomaly detection mechanism in step S5 specifically includes: S501, Real-time comparison and judgment: The amplitude, frequency, and energy parameters of the current sound signal are compared with the reference threshold during normal processing in real time. When the sound amplitude suddenly increases by more than 60dB above the preset threshold or the frequency characteristics change abruptly, it is judged as a level one abnormality. S502, Feature matching judgment: The spectral feature vector of the current sound signal is matched with the feature template in the fault audio sample library to calculate the similarity. The cosine similarity algorithm is used. When the similarity is ≥0.85, it is judged as a level 2 anomaly.

7. The method for monitoring and controlling abnormal noise from machine tool processing tools according to claim 6, characterized in that: The shutdown control triggering conditions in step S6 are: the duration of a first-level abnormality exceeds 10ms or a second-level abnormality is successfully matched; the shutdown command adopts a dual redundancy mechanism of hard-wired signals and communication commands.

8. The method for monitoring and controlling abnormal noise from machine tool processing tools according to claim 2, characterized in that: In step S7, sound data is automatically recorded and stored for 5 seconds before and after the anomaly occurs. At the same time, the monitoring function can be freely turned on and off by the user, and various parameter thresholds can be flexibly configured.

9. The method for monitoring and controlling abnormal noise from machine tool processing tools according to claim 1, characterized in that: In step S2, the following is adopted: MEMS microphone array, installed in the machine tool machining bay near the spindle; The central processing unit uses an ARM Cortex-M7 architecture microprocessor and runs a real-time operating system. The machine tool control interface supports multiple communication methods, including hard-wired I / O, Modbus bus, and OPC UA.

Images