Real-time intelligent monitoring and anomaly early warning method for machine tool status

By combining directional microphone arrays and beamforming technology, real-time intelligent monitoring of machine tool status is achieved, solving the problems of signal susceptibility to interference and slow response in existing technologies. It provides multi-level early warning and intelligent diagnostic capabilities, improving the accuracy and adaptability of the monitoring system.

CN120808817BActive Publication Date: 2026-11-17XIAMEN DINGYUN SOFTWARE
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202510926871.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-07-07
Publication Date
2026-11-17
Estimated Expiration
2045-07-07

AI Technical Summary

Technical Problem

Existing machine tool monitoring technologies are susceptible to environmental vibration interference, resulting in decreased signal quality, slow current signal response, inability to quickly capture machining anomalies, difficulty in accurately acquiring and processing acoustic signals, and inability to achieve multi-level early warning and intelligent diagnosis.

Method used

A directional microphone array is used to collect cutting acoustic signals. Beamforming technology is used to focus the cutting area. Combined with frequency separation, adaptive noise cancellation and multi-dimensional feature extraction, anomaly detection is performed based on dynamic matching of the CNC program, triggering multi-level early warnings.

Benefits of technology

It improves the stability and accuracy of signal acquisition, enhances the intelligence of signal processing, realizes a comprehensive description of the processing status and multi-level early warning, and improves the adaptability and intelligence level of the monitoring system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120808817B_ABST
    Figure CN120808817B_ABST
Patent Text Reader

Abstract

This invention discloses a method for real-time intelligent monitoring and anomaly early warning of machine tool status, comprising the following steps: acquiring cutting acoustic signals during machine tool machining using a directional microphone array; focusing the sound in the cutting area using beamforming technology and suppressing noise in non-target areas; performing frequency separation, noise reduction enhancement, and signal segmentation preprocessing on the acquired acoustic signals; extracting the time-domain, frequency-domain, and time-frequency features of the acoustic signals; dynamically matching the real-time acoustic features with a reference acoustic model based on the process steps in the machine tool CNC program; identifying deviations in acoustic features using an anomaly detection algorithm and triggering multi-level early warning signals according to the degree of deviation; demonstrating high creativity with unique signal acquisition and processing technology, dynamic matching mechanism, and intelligent early warning classification, surpassing the limitations of existing technologies, and significantly improving the accuracy, adaptability, and intelligence level of machine tool status monitoring systems.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing technology, specifically relating to a method for real-time intelligent monitoring and early warning of abnormalities in machine tool status. Background Technology

[0002] Monitoring the condition of machine tools during machining is crucial for ensuring machining quality, extending equipment lifespan, and preventing safety accidents. Current machine tool monitoring technologies primarily rely on the following methods:

[0003] Vibration signal monitoring:

[0004] Vibration sensors are often used to monitor tool wear, chipping, and the rigidity of machine tools.

[0005] Problem: Vibration signals are easily interfered with by vibrations from the machine tool base and the surrounding environment, leading to a decrease in signal quality. Vibration sensors are complex to install, requiring close proximity to critical machine tool components, resulting in high installation and maintenance costs.

[0006] Current signal monitoring:

[0007] By monitoring changes in the current of the spindle motor or feed motor, tool load and machining abnormalities can be indirectly determined.

[0008] Problem: The current signal response is slow, making it unable to quickly capture instantaneous anomalies during processing. The accuracy in identifying subtle processing conditions is low, especially under complex process steps or material processing conditions.

[0009] Because machining processes generate significant noise, acoustic signal monitoring is rarely used. Even if acoustic signal monitoring were considered, the following challenges would arise:

[0010] Signal Acquisition: How to accurately capture acoustic signature signals in the machine tool cutting area and effectively shield environmental noise in non-target areas?

[0011] Signal processing: How to enhance the target signal and remove irrelevant noise through dynamic adaptive processing methods?

[0012] Feature extraction: How to comprehensively extract multi-dimensional voiceprint features and accurately describe the processing status?

[0013] Anomaly monitoring: How can dynamic matching methods be used to improve the flexibility of monitoring and achieve intelligent anomaly diagnosis and multi-level early warning? Summary of the Invention

[0014] Therefore, the purpose of this invention is to provide a method for real-time intelligent monitoring and abnormal early warning of machine tool status. It demonstrates a high degree of creativity with its unique signal acquisition and processing technology, dynamic matching mechanism and intelligent early warning classification, which surpasses the limitations of existing technologies and significantly improves the accuracy, adaptability and intelligence level of machine tool status monitoring system.

[0015] To address the aforementioned technical problems, according to one aspect of the present invention, the present invention provides the following technical solution:

[0016] A method for real-time intelligent monitoring and anomaly warning of machine tool status includes the following steps:

[0017] A directional microphone array is used to collect cutting acoustic signals during machine tool processing;

[0018] Sound beamforming technology is used to focus sound in the cutting area and suppress noise in non-target areas;

[0019] The acquired voiceprint signals are subjected to frequency separation, noise reduction and enhancement, and signal segmentation preprocessing.

[0020] Extract the time-domain features, frequency-domain features, and time-frequency features of the voiceprint;

[0021] Based on the process steps in the machine tool CNC program, the real-time acoustic signature features are dynamically matched with the reference acoustic signature model;

[0022] Anomaly detection algorithms are used to identify deviations in voiceprint features and trigger multi-level warning signals based on the degree of deviation.

[0023] The directional microphone array is arranged in the key machining area of ​​the machine tool to cover the contact points between the spindle, the tool and the workpiece and their surrounding areas, to collect the sound signals in the area and to shield noise sources in other unrelated areas.

[0024] The beamforming technology uses the time delay and gain adjustment of the array microphones to form a directional focusing beam, which concentrates acoustic energy in the cutting area and suppresses background noise and environmental interference signals through beam filtering.

[0025] The preprocessing step includes:

[0026] A bandpass filter is used to separate the acoustic signal by frequency, retaining the main frequency band signal during the cutting process and eliminating background noise in the irrelevant frequency band.

[0027] The signal in the target area is further enhanced by an adaptive noise cancellation algorithm.

[0028] The voiceprint feature extraction includes the following:

[0029] Extract time-domain features, including the signal's peak value, root mean square value, and impulse factor;

[0030] Extract frequency domain features, including spectral energy distribution, dominant frequency shift, and spectral kurtosis;

[0031] The time-frequency variation characteristics of the voiceprint signal are extracted by short-time Fourier transform.

[0032] The dynamic matching is based on the process steps in the machine tool CNC program. A reference acoustic model is established for each process step, and the real-time acoustic signal is compared with the reference model of the corresponding process step.

[0033] The anomaly detection algorithm includes:

[0034] Calculate the Euclidean distance between the real-time voiceprint features and the baseline voiceprint model. If the Euclidean distance exceeds a set threshold, it is judged as abnormal.

[0035] The dynamic time warping algorithm is used to match the voiceprint time series and identify the cumulative change trend of the voiceprint.

[0036] The multi-level early warning signals include the following levels:

[0037] Level 1 warning: Slight deviation, prompting adjustment of processing parameters;

[0038] Level 2 warning: Significant deviation, it is recommended to stop the machine for inspection;

[0039] Level 3 warning: Severe anomaly, triggering automatic machine tool shutdown protection.

[0040] The anomaly detection results, combined with the physical correlation of voiceprint features, are used to classify and diagnose the causes of the anomalies, including:

[0041] The absence of high-frequency features is identified as tool wear.

[0042] Sudden changes in sound intensity are identified as tool chipping or machining of hard spots in the material.

[0043] Enhanced low-frequency characteristics indicate that the machine tool is loose or lacks rigidity.

[0044] An increase in full-spectrum signal energy indicates processing overload or material abnormality.

[0045] The voiceprint analysis results are presented through a visual interface, including real-time voiceprint spectra, anomaly classification prompts, and early warning log records, allowing operators to quickly determine the machine tool's operating status.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] 1. Innovative Acquisition Technology: Combination of Directional Microphone Array and Beamforming Technology

[0048] Existing technologies typically rely on a single microphone or a microphone array in a fixed position, making it difficult to track changes in the tool position in real time. However, this invention significantly improves the stability and accuracy of signal acquisition by combining a dynamically adjusted directional microphone array with beamforming technology.

[0049] Precise acquisition of cutting acoustic signals: A directional microphone array covers the spindle, tool-workpiece contact point, and surrounding area, significantly reducing environmental noise interference with the cutting acoustic signals. Dynamic adjustment of the microphone direction ensures that signal acquisition is always focused on the tool machining point, achieving a high signal-to-noise ratio even in complex machining environments.

[0050] High-sensitivity beamforming technology: Utilizing the time delay and gain adjustment of a microphone array to create a directional focused beam, acoustic energy is concentrated on the cutting area. By suppressing noise sources in non-target areas (such as machine tool vibration, coolant jet noise, etc.), the quality of sound acquisition in the target area is ensured.

[0051] Beamforming technology, based on the beam directionality control principle of microphone arrays, enhances the sound signal at the tool contact point by compensating for time delays and weighting the superposition of sound wave signals, while effectively suppressing noise from other directions. This technology, combined with optimized layout of the directional microphone array, solves the problem of traditional single-point microphones being susceptible to environmental noise, achieving precise acoustic capture of dynamic machining processes.

[0052] 2. Intelligent Enhancement of Signal Preprocessing: Frequency Separation and Adaptive Noise Cancellation

[0053] Traditional methods for signal filtering and denoising often use fixed parameters, making it difficult to adapt to dynamic changes in the processing environment. This invention enhances the intelligence and robustness of signal processing by combining bandpass filtering with adaptive noise cancellation.

[0054] Frequency separation: Using a bandpass filter to remove background noise in irrelevant frequency bands while retaining the main frequency bands of the cutting sound signal, thus significantly improving the effective information content of the signal.

[0055] Adaptive noise cancellation: By using an adaptive noise cancellation algorithm and combining it with real-time environmental noise reference, background noise is dynamically estimated and removed, thereby further enhancing the target signal.

[0056] Frequency separation is based on filtering theory, which decomposes the signal into different frequency components and eliminates interference signals in non-target frequency bands.

[0057] Adaptive noise cancellation enables the system to dynamically optimize the target signal as environmental noise changes by adjusting the noise model in real time.

[0058] 3. Comprehensiveness of multi-dimensional feature extraction

[0059] Compared with existing monitoring methods that rely on single or few features, this invention significantly improves the comprehensiveness and accuracy of the description of the processing state through multi-dimensional feature extraction, which helps to accurately identify complex abnormal patterns.

[0060] Rich feature dimensions: Voiceprint features are extracted comprehensively from the time domain, frequency domain, and time-frequency domain, fully reflecting the diversity of processing states.

[0061] Accuracy of feature description: Time-domain features quantify the intensity and stability of cutting sound; frequency-domain features reflect the frequency characteristics and distribution of the signal; time-frequency domain features describe the dynamic changes of the signal and the time-frequency distribution during the machining process.

[0062] Time-domain feature extraction relies on the direct analysis of signal amplitude; frequency-domain features are based on Fourier transform to convert time signals into frequency signals; time-frequency features capture the dynamic changes of sound signals during processing through short-time Fourier transform, revealing the complex patterns of the processing process.

[0063] 4. Dynamic matching and adaptive model based on CNC process steps

[0064] In existing technologies, monitoring systems typically use static models or fixed thresholds for anomaly detection, which cannot adapt to multiple process steps and time variations. This invention significantly enhances the adaptability and accuracy of the monitoring system through dynamic matching and time series analysis.

[0065] Dynamic matching: Based on the process steps of the CNC program, a reference acoustic fingerprint model is established for different machining processes, and the model is dynamically adjusted to adapt to changes in machining. Using real-time acoustic fingerprint features and dynamic matching with the reference model significantly improves the flexibility of status monitoring.

[0066] Time series matching: By using a dynamic time warping algorithm, non-linear changes in the time series during processing are allowed, thus solving the problem of the impact of processing rhythm fluctuations or time offsets on monitoring accuracy.

[0067] Dynamic matching, based on statistical learning theory, compares real-time features with a benchmark model formed by training data; the DTW algorithm optimizes the alignment of time series through dynamic programming, enabling the model to adapt to complex time biases.

[0068] 5. The practicality and intelligence of multi-level early warning and anomaly classification

[0069] Traditional monitoring systems often employ simple single-level early warning systems, failing to provide multi-level responses and anomaly classification. This invention, through multi-level early warning and classification diagnosis, endows the system with intelligent decision-making capabilities, providing operators with richer information and guidance.

[0070] Multi-level early warning mechanism: Triggering three-level early warning signals based on real-time characteristic deviations ensures a graded response from parameter adjustment to automatic shutdown, improving the efficiency and safety of anomaly handling.

[0071] Anomaly Classification and Diagnosis: By combining changes in voiceprint characteristics, accurate classification of various abnormal causes such as tool wear, chipping, machine tool loosening, and machining overload can be achieved.

[0072] Multi-level early warning is based on feature deviation and time series matching results to manage complex processing states in a hierarchical manner; anomaly classification relies on the physical meaning of voiceprint features and is combined with processing technology knowledge for diagnosis. Attached Figure Description

[0073] To more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and detailed embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0074] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0075] Figure 2 This is a flowchart illustrating step 1) of the method of the present invention;

[0076] Figure 3 This is a flowchart illustrating step 2) of the method of the present invention;

[0077] Figure 4 This is a flowchart illustrating step 3) of the method of the present invention;

[0078] Figure 5 This is a flowchart illustrating step 4) of the method of the present invention;

[0079] Figure 6 This is a flowchart illustrating step 5) of the method of the present invention;

[0080] Figure 7 This is a flowchart illustrating step 6) of the method of the present invention. Detailed Implementation

[0081] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0082] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0083] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0084] See Figure 1 The method for real-time intelligent monitoring and anomaly warning of machine tool status includes the following steps:

[0085] 1) See Figure 2 This method utilizes a directional microphone array to collect cutting acoustic signals during machine tool processing. The array is positioned in key machining areas of the machine tool to cover the spindle, the contact point between the tool and the workpiece, and their surrounding area, collecting sound signals from these areas while shielding noise sources from other irrelevant areas. Each microphone in the array is mounted on an independent rotatable bracket, which includes a servo motor or stepper motor to achieve horizontal and vertical rotation. The horizontal rotation angle range is -45° to +45°, and the vertical rotation angle range is also -45° to +45°. The microphone rotation angle is adjusted according to the real-time position of the machine tool to ensure that the beamforming technology always focuses on the contact point between the tool and the workpiece.

[0086] The specific implementation method is as follows:

[0087] 1. Real-time tool position acquisition: Obtain the real-time tool position r from the CNC system. s (t)=(x s (t),y s (t),z s (t)). The actual position of the tool is calibrated using industrial camera-assisted monitoring, especially during non-linear path machining. Sampling frequency: 1kHz (1000 samples per second) ensures real-time response to changes in the machining path.

[0088] 2. Target area definition: The contact point between the tool and the workpiece is the target area, the range of which is the tool diameter or the radius R of the machining range. d Typical values ​​range from 5mm to 50mm.

[0089] 3. Microphone rotation angle calculation: For each microphone position r m =(x m ,y m ,z m ), calculate the target region r s The relative direction of (t): horizontal angle θ h and vertical angle θ v .

[0090] 4. Adjust the rotation angle of the servo motor or stepper motor in real time according to the calculation results so that the microphone is always pointing at the target area.

[0091] 5. Dynamic delay calculation: Due to the continuous change in tool position during machining, the distance d from the sound source to the microphone... m(t) also changes. Sound propagation delay τ m (t) Dynamic updates, expressed as:

[0092] d m (t)=||r m -r s (t)||;where, τ m (t): Sound propagation delay (seconds): The time delay for sound to travel from the target area to the m-th microphone. d m (t): Distance between the target area and the microphone (meters, m): Spatial distance between the target area and the m-th microphone. m The position vector (m, m) of the m-th microphone: the three-dimensional spatial position of the microphone in the array coordinate system, expressed as (x... m ,y m ,z m ). r s (t): Real-time coordinate vector (m, m) of the target area (tool position): Real-time position of the tool-workpiece contact point in three-dimensional space, expressed as (x s (t),y s (t),z s (t)). c: Speed ​​of sound (m / s): The speed at which sound waves travel in the air, 340-345 m / s (corrected for ambient temperature).

[0093] 2) See Figure 3 The beamforming technique focuses the sound in the cutting area and suppresses noise in non-target areas. The beamforming technique forms a directional focusing beam by adjusting the time delay and gain of the array microphones, concentrating acoustic energy in the cutting area, and suppressing background noise and environmental interference signals through beam filtering.

[0094] The specific implementation method is as follows:

[0095] 1. Delay compensation and signal superposition: For the signal x acquired by each microphone m (t), based on the dynamic delay τ m (t) Compensation time:

[0096] x m ′(t)=x m (t+τ m (t)); where x m (t): The raw signal received by the microphone: the time-domain signal directly acquired by the m-th microphone, including the target sound signal and noise. x′ m (t): The signal after time delay compensation: the signal aligned by time delay compensation. τ m(t): Sound propagation delay (seconds): Same as above, a parameter used for time alignment of the signal.

[0097] 2. A focused signal is formed through dynamic weighted superposition:

[0098] Where y(t): focused sound beam output signal; M: number of microphones: enhanced signal focused on the target area, obtained by superimposing multi-channel signals. m (t): Weighting factor for the m-th microphone: a weighting factor used to dynamically adjust the contribution of signal superposition; if a delay and superposition method is used: If the MVDR method is used: dynamic optimization calculations are performed to minimize noise. x′ m (t): Delay-compensated signal: The m-th microphone signal after delay compensation, used to superimpose to form a focused sound beam signal.

[0099] 3. Noise suppression:

[0100] Use filters to enhance the output signal:

[0101] y′(t)=H(f)·y(t); where, y′(t): the filtered output signal. H(f): the filter frequency response: enhances the target frequency band signal and suppresses non-target frequency band noise.

[0102] 3) See Figure 4 The acquired acoustic signature signal is subjected to frequency separation, noise reduction and enhancement, and signal segmentation preprocessing. The signal segmentation preprocessing step includes: using a bandpass filter to separate the acoustic signature signal by frequency, retaining the main frequency band signal during the cutting process, and removing background noise in non-relevant frequency bands; and further enhancing the signal in the target area through an adaptive noise cancellation algorithm.

[0103] The specific implementation method is as follows:

[0104] 1. Frequency separation: The target frequency band (cutting sound frequency range 3kHz-15kHz) signal is retained by a bandpass filter, and background noise of irrelevant frequency bands is removed.

[0105] Bandpass filter design: Using IIR (Infinite Impulse Response) or FIR (Finite Impulse Response) filters to achieve the signal's frequency band:

[0106] x f (t)=BPF(x(t),f low ,f high x(t): the original time-domain input signal. f (t): The filtered signal retains only the components of the target frequency band. BPF(): Bandpass filter function. f lowThe low-frequency cutoff point of the filter. high : The high-frequency cutoff point of the filter.

[0107] 2. Noise reduction and enhancement

[0108] The target signal is further enhanced and the remaining background noise is reduced by using an adaptive noise cancellation algorithm (LMS algorithm).

[0109] Noise model: Where, x clean (t): The denoised signal. x f (t): The signal after frequency separation. Estimated background noise signal.

[0110] Noise estimation (LMS algorithm): This method uses an adaptive filter to estimate and eliminate background noise. in, The estimated background noise signal. w(t): the weight vector of the adaptive filter. z(t): the vector of the input reference noise signal.

[0111] Weight Update Rule (LMS): w(t+1) = w(t) + μ·e(t)·z(t); where, w(t): the weight vector of the filter at the current time step; μ: step size parameter, controlling the convergence speed and stability, typically 0.01 to 0.1; e(t): error signal, representing the difference between the denoised signal and the reference noise.

[0112] 3. Signal segmentation preprocessing

[0113] Segmenting long-term signals facilitates subsequent feature extraction.

[0114] Segment length: L: 50-100 milliseconds (ms), depending on the characteristics of the cutting acoustic signal.

[0115] Number of sampling points in each segment: N = L·f s N: Number of sampling points per signal segment. L: Segment duration (seconds). f s Sampling frequency (Hertz, Hz).

[0116] Segment overlap: To reduce the breakpoint effect, the overlap rate R between adjacent segments is 50%-75%.

[0117] Piecewise matrix:

[0118] Where X: a segmented signal matrix, with each row representing a segmented signal. k [1:N]: Sample value of the k-th segment of the signal.

[0119] 4) See Figure 5 The process involves extracting the time-domain, frequency-domain, and time-frequency features of the voiceprint. Voiceprint feature extraction includes the following: extracting time-domain features, including the peak value, root mean square value, and impulse factor of the signal; extracting frequency-domain features, including spectral energy distribution, dominant frequency shift, and spectral kurtosis; and extracting the time-frequency variation features of the voiceprint signal through short-time Fourier transform.

[0120] The specific implementation method is as follows:

[0121] 1. Extract target features:

[0122] Peak value: P = max(x(t)); where, P: the maximum amplitude of the signal: reflecting the instantaneous maximum value of the signal, used to measure the signal strength. x(t): the time domain representation of the signal: the amplitude of the cutting acoustic signal at a certain time t.

[0123] Root mean square (RMS): Where, RMS: Root Mean Square value of the signal: measures the energy of the signal and reflects the strength of the power signal during the cutting process. N: Number of signal sampling points: the total number of signal samples in the current time period. x 2 (t): The square of the signal at time t.

[0124] Crest Factor (CF): Among them, CF: impulse factor: reflects the peak characteristics of the signal waveform; the higher the value, the greater the signal fluctuation.

[0125] 2. Frequency Domain Feature Extraction

[0126] Power spectral density (PSD): The power spectral density of a signal is calculated using a Fast Fourier Transform (FFT). Wherein, PSD(f): power spectral density (V) at frequency f. 2 / Hz): Represents the energy distribution of the signal at different frequencies. X(f): Frequency domain representation of the signal, obtained through Fast Fourier Transform. f: Frequency (Hertz, Hz): The frequency component of the signal. x(t): The amplitude of the signal in the time domain at time t.

[0127] Main frequency offset: Among them, f dom Main frequency (Hertz, Hz): The frequency corresponding to the peak value of the power spectral density, used to determine changes in the frequency characteristics of a signal.

[0128] Spectral kurtosis (SK): Wherein, SK: spectral kurtosis: used to describe the degree of concentration of the spectrum, a higher value indicates that the spectrum is concentrated in a certain frequency range. E[·]: mathematical expectation operator: calculates the mean and fluctuation of the power spectrum distribution.

[0129] 3. Time-frequency feature extraction

[0130] Short-Time Fourier Transform (STFT): Analyzes the dynamic changes of a signal in time and frequency using the short-time Fourier transform: X(f,t)=∫x(τ)w(τ-t)e -j2πfτ dτ; where X(f,t): complex values ​​of frequency f and time t in the time-frequency plot, representing the amplitude of the signal at time t and frequency f. w(τ-t): window function: used to segment the signal and analyze local characteristics; window functions include the Hamming window and the Hann window.

[0131] Time-frequency energy distribution: By squared-ratio of the STFT amplitude, the time-frequency energy distribution is obtained: E(f,t)=|X(f,t)| 2 Wherein, E(f,t): time-frequency energy distribution: represents the energy of the frequency components at different time points.

[0132] 5) See Figure 6 Based on the process steps in the machine tool CNC program, the real-time acoustic features are dynamically matched with the reference acoustic model. The dynamic matching is based on the process steps in the machine tool CNC program. Reference acoustic models are established for different process steps, and the real-time acoustic signals are compared with the reference models of the corresponding process steps.

[0133] The specific implementation method is as follows:

[0134] 1. Construction of the baseline voiceprint model

[0135] Feature extraction: For each process step, extract specific voiceprint features: F baseline =[P,RMS,CF,f dom ,SK];wherein, F baseline Reference acoustic signature feature vector: Used to describe the acoustic characteristics of a certain process step under normal operating conditions. P, RMS, CF, f dom SK: peak value, root mean square value, impact factor, dominant frequency, and spectral kurtosis, respectively (see step 4).

[0136] Process step number: Each process step i corresponds to a baseline model: F baseline,i Where i: the process step number (no unit).

[0137] 2. Real-time voiceprint feature extraction

[0138] Real-time acquisition of cutting sound signals during the current machining process, extraction of their feature values, and formation of a real-time feature vector: F real=[P′,RMS′,CF′,f′] dom ,SK′];wherein, F real Real-time voiceprint feature vector: The feature set of the sound signal under the current operating conditions. P′,RMS′,CF′,f′ dom SK′: Real-time extracted peak value, root mean square value, impact factor, dominant frequency and spectral kurtosis.

[0139] 3. Dynamic matching

[0140] Feature deviation calculation: Calculate the deviation between real-time voiceprint features and the baseline voiceprint model: Δ=||F real -F baseline ||;Where Δ: Feature bias: describes the magnitude of the difference between real-time features and baseline features.

[0141] formula: Among them, F real,k : The k-th term of the real-time voiceprint features. F baseline,k : The k-th term of the baseline voiceprint feature. n: The dimension of the feature.

[0142] Dynamic Time Warping (DTW) performs dynamic matching at the time series level, allowing for non-linear deformation of the time axis: DTW(F real ,F baseline ); where DTW: the output value of the Dynamic Time Warping algorithm: used to calculate the minimum cumulative distance between two time series. Implementation: solved through dynamic programming. Output value: the similarity between the two sequences; the smaller the value, the closer they are.

[0143] 4. Dynamic determination of matching results

[0144] Deviation threshold determination: Based on the characteristic deviation Δ and the dynamic time warp distance DTW, anomaly detection conditions are set.

[0145] Normal: Δ≤Δ thresh And DTW≤DTW thresh .

[0146] Abnormal: Otherwise, it is judged as abnormal.

[0147] Where, Δ thresh Threshold for feature bias. DTW thresh DTW threshold, typical value is determined based on the training model.

[0148] The threshold is dynamically adjusted according to different process steps i: Δ thresh,i DTW thresh,i Where i: the current process step number.

[0149] 6) See Figure 7The system identifies deviations in voiceprint features using an anomaly detection algorithm and triggers multi-level warning signals based on the degree of deviation. The anomaly detection algorithm includes: calculating the Euclidean distance between real-time voiceprint features and a benchmark voiceprint model; if the Euclidean distance exceeds a set threshold, it is determined to be an anomaly; and using a dynamic time warping algorithm to match the voiceprint time series and identify the cumulative change trend of the voiceprint.

[0150] Multi-level warning signals include the following levels:

[0151] Level 1 warning: Slight deviation, prompting adjustment of processing parameters;

[0152] Level 2 warning: Significant deviation, it is recommended to stop the machine for inspection;

[0153] Level 3 warning: Severe anomaly, triggering automatic machine tool shutdown protection.

[0154] Anomaly detection results are combined with the physical correlation of voiceprint features to classify and diagnose the causes of anomalies, including:

[0155] The absence of high-frequency features is identified as tool wear.

[0156] Sudden changes in sound intensity are identified as tool chipping or machining of hard spots in the material.

[0157] Enhanced low-frequency characteristics indicate that the machine tool is loose or lacks rigidity.

[0158] An increase in full-spectrum signal energy indicates processing overload or material abnormality.

[0159] Furthermore, the voiceprint analysis results are presented through a visual interface, including real-time voiceprint spectra, anomaly classification prompts, and early warning log records, allowing operators to quickly determine the machine tool's operating status.

[0160] The present invention provides specific embodiments as follows:

[0161] Specific Implementation Example 1: Monitoring of Cutting Sounds in Aluminum Alloy Parts Machining

[0162] In high-speed machining of aluminum alloy parts, due to the softness of the material and the high machining speed, the wear condition of the cutting tool has a significant impact on cutting noise. This system is used to monitor the tool condition in real time during aluminum alloy cutting and to detect abnormalities.

[0163] Implementation steps:

[0164] 1. Microphone array configuration:

[0165] Sixteen microphones were arranged in a rectangular array, with the center of the array 100mm away from the tool processing area.

[0166] Each microphone is mounted on an independent stand and can rotate ±45° in both the horizontal and vertical directions to track the tool position in real time.

[0167] 2. Beamforming and noise suppression:

[0168] The system focuses on the contact point between the tool and the aluminum alloy, shielding the machine tool spindle noise and coolant jet noise.

[0169] The target frequency band signal of 3kHz-12kHz is preserved by using a bandpass filter.

[0170] 3. Feature extraction and dynamic matching:

[0171] Peak value, RMS, impact factor, main frequency, and spectral kurtosis are extracted as real-time features.

[0172] Dynamically match the baseline voiceprint model corresponding to the CNC program, and calculate the feature deviation and time series matching distance.

[0173] 4. Anomaly detection and early warning:

[0174] Event 1: Tool wear caused a drop in the main frequency. The system detected that the main frequency was below 5kHz (reference range: 6.5-7.5kHz), triggering a level 1 warning.

[0175] Event 2: Tool chipping caused a sudden increase in peak value and RMS, with the peak value reaching twice the baseline value, triggering a level 3 warning and automatic shutdown.

[0176] Specific Implementation Example 2: Monitoring Cutting Sounds in Rough Steel Machining

[0177] In the rough machining of steel, the cutting forces are relatively large, which can easily lead to abnormalities such as tool breakage or material hardening. This system is used to monitor the cutting sound signals of steel in real time and detect machining overload and tool breakage problems.

[0178] Implementation steps:

[0179] 1. Microphone array configuration:

[0180] Twenty-four microphones are arranged in a hemispherical array to cover the area around the contact points between the spindle, tool, and workpiece.

[0181] The array has a radius of 120mm and is installed inside a protective cover in the machine tool processing area.

[0182] 2. Beamforming and frequency separation:

[0183] It focuses on the cutting sound signal at the contact point between the tool and the workpiece, and suppresses low-frequency noise generated by machine tool vibration.

[0184] The cutting sound frequency band of 5kHz-15kHz is preserved by using a filter.

[0185] 3. Feature extraction and dynamic matching:

[0186] The system extracts the peak value, RMS, main frequency offset, and time-frequency energy distribution of the cutting sound.

[0187] Dynamically match real-time features with the benchmark model to detect feature bias and cumulative change trends.

[0188] 4. Anomaly detection and early warning:

[0189] Event 1: RMS was detected to be continuously increasing and exceeding 1.5 times the baseline value, which was identified as processing overload, triggering a level 2 warning. It is recommended to adjust the processing parameters.

[0190] Event 2: The machining of material hard spots caused a sudden increase in the intensity and spectral energy of the cutting sound. The system identified the anomaly, triggered a level 3 warning, and shut down for protection.

[0191] Specific Implementation Example 3: Monitoring of Cutting Sounds in Titanium Alloy Finishing

[0192] The finishing of titanium alloys places high demands on tool condition; chipping or machining vibrations can significantly affect the surface quality of the parts. This system is used to accurately monitor the acoustic characteristics of titanium alloy machining and provide early warnings of tool problems.

[0193] Implementation steps:

[0194] 1. Microphone array configuration:

[0195] Use 32 microphones arranged in a double-layer planar array (evenly distributed in two layers, 16 microphones in each layer), 80mm away from the tool area.

[0196] The upper and lower layers of the array respectively collect sound signals from the processing area to form a three-dimensional sound beam focusing.

[0197] 2. Beamforming and noise reduction enhancement:

[0198] The system simultaneously focuses on the contact point between the tool and the workpiece and its surrounding area to suppress high-frequency noise that may be generated during titanium alloy cutting.

[0199] An adaptive noise cancellation algorithm combined with a bandpass filter enhances the cutting sound signal in the 8kHz-15kHz frequency band.

[0200] 3. Feature extraction and dynamic matching:

[0201] Extract the peak value, impact factor, main frequency offset, and time-frequency energy distribution of the cutting sound.

[0202] STFT was used to generate time-frequency diagrams to analyze the time-frequency characteristics of the acoustic signal from titanium alloy cutting and to dynamically match the reference acoustic pattern model.

[0203] 4. Anomaly detection and early warning:

[0204] Event 1: Tool wear caused loss of high-frequency characteristics, and the main frequency dropped to 9kHz (reference range: 10-12kHz), triggering a level 1 warning.

[0205] Event 2: Increased processing vibration led to enhanced low-frequency components and a significant shift in time-frequency energy distribution, triggering a level-two warning.

[0206] Event 3: Tool chipping caused the signal peak value to reach twice the normal value. The system identified this as a serious anomaly, triggered a level 3 warning, and automatically shut down for protection.

[0207] Summarize

[0208] These three examples demonstrate the system's adaptability to different processing materials and operating conditions:

[0209] Aluminum alloy machining: lightweight cutting, with a focus on detecting tool wear and chipping.

[0210] Rough machining of steel: high-load cutting, with a focus on identifying machining overload and hard spot problems.

[0211] Titanium alloy finishing: High precision requirements, with a focus on monitoring tool vibration and high-frequency characteristic changes.

[0212] Based on the above embodiments, in order to further improve the system's ability to identify atypical anomalies under complex process conditions, this embodiment introduces a process context awareness engine, which incorporates semantic information in the processing into the voiceprint feature judgment process, so as to realize the semantic consistency verification between voiceprint features and actual processing conditions, and ultimately achieve a more accurate anomaly warning and a more interpretable anomaly judgment mechanism.

[0213] This engine is integrated into the voiceprint anomaly detection system in a modular manner. Its core idea is no longer limited to the traditional "signal exceeding threshold alarm" mode, but rather establishes a semantic mapping relationship between "process state - voiceprint expectation" so that the system can identify whether the current voiceprint is semantically "reasonable". In short, the system not only judges whether the signal strength is abnormal, but also judges "whether such a signal should appear in such process state".

[0214] The entire processing flow includes the following steps:

[0215] First, the system extracts the process parameters of the current machining segment in real time through the communication interface with the machine tool's CNC controller. These parameters include the program segment number, tool number and type, spindle speed (RPM), feed rate, workpiece material type, cooling status, and tool change status. This information is encapsulated into a structured set of semantic tags to represent the behavioral semantics of the machining context.

[0216] Subsequently, the system constructs a process semantic graph based on historical normal processing data, mapping semantic labels to semantic vectors and clustering them into multiple representative process state nodes. Each node corresponds to a typical processing semantic state and is associated with a set of "expected voiceprint feature ranges" obtained based on statistical analysis, such as the reasonable range of the main frequency, the typical value of RMS, and the upper limit of the impact factor. The directed edges in the graph represent the logical transition paths between different processing states, such as from roughing to finishing, from the main processing section to the tool change section, so that the system can identify the rationality of semantic changes.

[0217] In actual operation, the system locates the most similar node in the graph based on the current semantic label and extracts the expected voiceprint feature set of that node as the "semantic target voiceprint model" for the current processing state. Subsequently, the system compares the voiceprint features extracted in real time with this model and calculates the "semantic deviation" index to determine whether the current voiceprint matches the current process state.

[0218] The deviation is composed of two parts: the first part is the Euclidean distance between the voiceprint features and the semantic target model, used to quantify the numerical deviation of the signal; the second part is the logical contradiction score between the semantic label and the voiceprint behavior, such as the appearance of typical tool-changing impact sounds during non-tool-changing stages, or the appearance of roughing characteristics during low-speed finishing. The system weights and sums the two parts to obtain a unified semantic deviation, and triggers anomaly warnings of the corresponding level based on this.

[0219] When the deviation is within the normal range, the system considers the machining status reasonable and no alarm is needed. When the deviation exceeds the set threshold, the system will trigger a level 1, 2, or 3 warning, corresponding to a response mechanism of suggested parameter adjustment, pause inspection, or automatic shutdown protection, respectively. Simultaneously, based on the logical determination of the anomaly source, the system generates natural language output with semantic interpretation capabilities, such as: "The current process status is low-speed finishing. A continuous impact-type high-frequency signal has been detected, suspected to be tool micro-breakage or abnormal feed vibration."

[0220] The introduction of this context-aware engine essentially expands anomaly detection from a single signal feature comparison to a dual rationality judgment mechanism of "acoustic performance - process semantics." The fundamental principle behind its effectiveness lies in establishing an analytical paradigm shift from "signal deviation" to "semantic rationality verification." The system not only detects "whether the value is abnormal," but also determines "whether the anomaly is semantically valid."

[0221] In the field verification under low-speed precision machining conditions of titanium alloys, traditional acoustic signature methods mistakenly identified the high-frequency fluctuations of "feed micro-vibration excitation" as tool change impact. However, this embodiment, by analyzing the current semantic tags and finding no tool change action, combined with differences in spectral characteristics, successfully identified it as a tool micro-breakage anomaly and triggered a level-two warning. The system achieves high-reliability identification of atypical anomalies and outputs diagnostically targeted explanatory statements.

[0222] After multi-scenario comparative testing, the introduction of the process context awareness engine improved the system's recognition accuracy in atypical abnormal scenarios from 82.7% to 93.4%, reduced the false alarm rate by about 48%, and achieved an intelligent explanation output accuracy of over 90%, significantly enhancing the system's intelligence level and practical value.

[0223] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for real-time intelligent monitoring and anomaly early warning of machine tool status, characterized in that, Includes the following steps: A directional microphone array is used to collect cutting acoustic signals during machine tool processing; Sound beamforming technology is used to focus sound in the cutting area and suppress noise in non-target areas; The acquired voiceprint signals are subjected to frequency separation, noise reduction and enhancement, and signal segmentation preprocessing. Extract the time-domain features, frequency-domain features, and time-frequency features of the voiceprint; Based on the process steps in the machine tool CNC program, real-time voiceprint features are dynamically matched with a reference voiceprint model; a process context awareness engine is introduced to incorporate semantic information during the machining process into the voiceprint feature judgment process to achieve semantic consistency verification between voiceprint features and actual machining conditions; through the communication interface with the machine tool CNC controller, the process parameters of the current machining segment are extracted in real time; the process parameters are encapsulated into a structured set of semantic tags to characterize the behavioral semantics of the machining context. A process semantic graph is constructed based on historical normal processing data. Semantic labels are mapped to semantic vectors and clustered into multiple representative process state nodes. Each node corresponds to a typical processing semantic state and is associated with a set of expected voiceprint feature ranges obtained based on statistical analysis. Directed edges in the graph represent logical transition paths between different processing states, enabling the system to identify the rationality of semantic changes. Based on the current semantic label, the most similar node in the graph is located, and the expected voiceprint feature set of that node is extracted as the semantic target voiceprint model for the current processing state. The real-time extracted voiceprint features are compared with this model to calculate a semantic deviation index, which is used to determine whether the current voiceprint matches the current process state. This deviation consists of two parts: the first part is the Euclidean distance between the voiceprint features and the semantic target model, used to quantify the numerical deviation of the signal; the second part is the logical contradiction score between the semantic label and the voiceprint behavior. The system weights and sums the two results to obtain a unified semantic deviation, and triggers anomaly warnings of the corresponding level accordingly. Anomaly detection algorithms are used to identify deviations in voiceprint features and trigger multi-level warning signals based on the degree of deviation.

2. The method for real-time intelligent monitoring and anomaly early warning of machine tool status according to claim 1, characterized in that, The directional microphone array is arranged in the key machining area of ​​the machine tool to cover the spindle, the contact point between the tool and the workpiece and its surrounding area, to collect the sound signal of the area and to shield noise sources in other unrelated areas.

3. The method for real-time intelligent monitoring and anomaly early warning of machine tool status according to claim 1, characterized in that, The beamforming technology uses the time delay and gain adjustment of the array microphones to form a directional focused beam, concentrating acoustic energy in the cutting area, and suppressing background noise and environmental interference signals through beam filtering.

4. The method for real-time intelligent monitoring and anomaly early warning of machine tool status according to claim 1, characterized in that, The signal segmentation preprocessing step includes: A bandpass filter is used to separate the acoustic signal by frequency, retaining the main frequency band signal during the cutting process and eliminating background noise in the irrelevant frequency band. The signal in the target area is further enhanced by an adaptive noise cancellation algorithm.

5. The method for real-time intelligent monitoring and anomaly early warning of machine tool status according to claim 1, characterized in that, Voiceprint feature extraction includes the following: Extract time-domain features, including the signal's peak value, root mean square value, and impulse factor; Extract frequency domain features, including spectral energy distribution, dominant frequency shift, and spectral kurtosis; The time-frequency variation characteristics of the voiceprint signal are extracted by short-time Fourier transform.

6. The method for real-time intelligent monitoring and anomaly early warning of machine tool status according to claim 1, characterized in that, The anomaly detection algorithm includes: Calculate the Euclidean distance between the real-time voiceprint features and the baseline voiceprint model. If the Euclidean distance exceeds a set threshold, it is judged as abnormal. The dynamic time warping algorithm is used to match the voiceprint time series and identify the cumulative change trend of the voiceprint.

7. The method for real-time intelligent monitoring and anomaly early warning of machine tool status according to claim 1, characterized in that, The multi-level early warning signal includes the following levels: Level 1 warning: Slight deviation, prompting adjustment of processing parameters; Level 2 warning: Significant deviation, it is recommended to stop the machine for inspection; Level 3 warning: Severe anomaly, triggering automatic machine tool shutdown protection.

8. The method for real-time intelligent monitoring and anomaly early warning of machine tool status according to claim 1, characterized in that, Anomaly detection results are combined with the physical correlation of voiceprint features to classify and diagnose the causes of anomalies, including: The absence of high-frequency features is identified as tool wear. Sudden changes in sound intensity are identified as tool chipping or machining of hard spots in the material. Enhanced low-frequency characteristics indicate that the machine tool is loose or lacks rigidity. An increase in full-spectrum signal energy is determined to be due to processing overload or material abnormality.

9. The method for real-time intelligent monitoring and anomaly early warning of machine tool status according to claim 1, characterized in that, The voiceprint analysis results are presented through a visual interface, including real-time voiceprint spectra, anomaly classification prompts, and early warning log records, allowing operators to quickly determine the machine tool's operating status.

Citation Information

Patent Citations

  • Cutter state voiceprint detection method, process distinguishing method, detection system and machine tool

    CN118559500A