Terminal machine on-line analysis early warning method and device and related assembly

Abstract

The invention discloses a terminal machine online analysis early warning method and device and related components, and the method comprises the steps: collecting sound data and vibration data of a terminal machine in an operation process, and obtaining sound detection data and vibration detection data; preprocessing the sound detection data and the vibration detection data to obtain a detection time domain signal; converting the detection time domain signal into a detection frequency domain signal through fast Fourier transform; extracting detection characteristic data reflecting the operation state of the terminal machine from the detection frequency domain signal; matching the detection feature data with a preset fault feature model; and if the detection feature data is successfully matched with the preset fault feature model, determining that the terminal machine has a fault and sending a fault early warning signal. By collecting and analyzing the sound data and the vibration data in real time, potential faults of the terminal machine can be found in time, the downtime of the terminal machine is shortened, and the usability and reliability of the terminal machine are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wire harness generation, and in particular to an online analysis and early warning method and device for a terminal machine and related components. Background Art

[0002] A crimping machine is a type of wire harness production equipment widely used in the electronics and electrical industries. Its core function is to reliably connect wires to terminals through a crimping process, enabling circuit continuity and signal transmission. With the advancement of industrial automation, the production efficiency and process precision of crimping machines have significantly improved. However, crimping machines still face numerous challenges in terms of equipment status monitoring and product quality control.

[0003] In the production process of existing terminal machines, equipment status monitoring and quality control mainly rely on manual inspections and post-inspections. However, this approach has the following significant drawbacks:

[0004] Traditional manual inspections rely on the operator's experience and subjective judgment, making it difficult to understand the equipment's operating status in real time. Early signs of equipment failure are often overlooked, leading to delayed fault detection and increased downtime and repair costs. Summary of the Invention

[0005] The purpose of the present invention is to provide a terminal machine online analysis and early warning method, device and related components, aiming to solve the problem of delayed fault detection of the terminal machine.

[0006] In a first aspect, an embodiment of the present invention provides an online analysis and early warning method for a terminal machine, comprising:

[0007] Collect the sound data and vibration data of the terminal machine during operation to obtain sound detection data and vibration detection data;

[0008] Preprocessing the sound detection data and the vibration detection data to obtain a detection time domain signal;

[0009] Converting the detection time domain signal into a detection frequency domain signal through fast Fourier transform;

[0010] Extracting detection feature data reflecting the terminal machine operating state from the detection frequency domain signal;

[0011] Matching the detection feature data with a preset fault feature model;

[0012] If the detection feature data successfully matches the preset fault feature model, it is determined that the terminal machine has a fault and a fault warning signal is issued.

[0013] In a second aspect, an embodiment of the present invention provides an online analysis and early warning device for a terminal machine, comprising:

[0014] The acquisition unit is used to collect the sound data and vibration data of the terminal machine during operation to obtain the sound detection data and vibration detection data;

[0015] A preprocessing unit, configured to preprocess the sound detection data and the vibration detection data to obtain a detection time domain signal;

[0016] A conversion unit, configured to convert the detection time domain signal into a detection frequency domain signal through a fast Fourier transform;

[0017] An extraction unit, configured to extract detection feature data reflecting the terminal machine's operating status from the detection frequency domain signal;

[0018] A matching unit, configured to match the detection feature data with a preset fault feature model;

[0019] The determination unit is configured to determine that a fault exists in the terminal machine and issue a fault warning signal if the detection feature data successfully matches a preset fault feature model.

[0020] In a third aspect, an embodiment of the present invention further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the terminal machine online analysis and early warning method described in the first aspect is implemented.

[0021] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the terminal machine online analysis and early warning method described in the first aspect is implemented.

[0022] The present invention discloses a method, device, and related components for online analysis and early warning of a terminal machine. The method comprises: collecting sound and vibration data during the operation of the terminal machine to obtain sound detection data and vibration detection data; preprocessing the sound and vibration detection data to obtain a detection time domain signal; converting the detection time domain signal into a detection frequency domain signal through a fast Fourier transform; extracting detection feature data reflecting the terminal machine's operating status from the detection frequency domain signal; matching the detection feature data with a preset fault feature model; and if the detection feature data successfully matches the preset fault feature model, determining that the terminal machine has a fault and issuing a fault early warning signal. By collecting and analyzing sound and vibration data in real time, the present invention can promptly detect potential faults in the terminal machine, reducing the terminal machine's downtime and improving the terminal machine's availability and reliability. It also reduces the workload of manual inspections and testing, and improves the automation and intelligence level of the production process. The embodiments of the present invention also provide an online analysis and early warning device for a terminal machine, a computer-readable storage medium, and a computer device, all of which have the aforementioned beneficial effects and are not further described here. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 This is a flow chart of the online analysis and early warning method for the terminal machine;

[0025] Figure 2 This is a schematic block diagram of the terminal machine online analysis and early warning device. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0027] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0028] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0029] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0030] See also Figure 1 This embodiment provides a terminal machine online analysis and early warning method, including:

[0031] S101: collecting sound data and vibration data of the terminal machine during operation to obtain sound detection data and vibration detection data;

[0032] During the operation of the terminal machine, vibration data and sound data with a frequency range of 1-50kHz are collected respectively through vibration sensors (such as piezoelectric vibration sensors) and sound sensors (such as high-sensitivity microphone arrays) installed at key parts of the equipment.

[0033] S102: Preprocessing the sound detection data and the vibration detection data to obtain a detection time domain signal;

[0034] Specifically, the collected sound detection data and vibration detection data are preprocessed by filtering, denoising and other preprocessing operations to remove interference signals, and then the preprocessed sound detection data and vibration detection data are integrated to obtain a detection time domain signal.

[0035] More specifically, the vibration sensor's output signal is amplified by a range switching circuit to a preset range (e.g., 0.25 to 250 times), then pre-processed by an anti-aliasing low-pass filter (10th-order linear phase) to eliminate high-frequency noise. The analog signal is then converted to a digital signal by a 16-bit analog-to-digital converter (ADC) at a 1MHz sampling rate. Sound data is collected by a microphone array, filtered through a bandpass filter (1-50kHz) to remove ambient noise, and digitized by the ADC at a 48kHz sampling rate.

[0036] The processing of vibration data further includes: extracting the acceleration signal through software high-pass filtering, and obtaining the velocity signal and displacement signal through single integration and double integration respectively, so as to restore the dynamic characteristics of the equipment operation. Subsequently, the three types of signals are subjected to software low-pass filtering to obtain the acceleration high-frequency filter signal, low-frequency filter signal, velocity filter signal and displacement filter signal. For different frequency components, a differentiated resampling strategy is adopted: the acceleration high-frequency signal is resampled at 51.2kHz, the low-frequency signal is resampled at 12.8kHz, the velocity signal is resampled at 2.56kHz, and the displacement signal is resampled at 1.28kHz. Finally, the acceleration high-frequency signal is converted into a low-frequency envelope signal of 0-1kHz through envelope demodulation, and a fast Fourier transform (FFT) is performed together with the remaining signals to generate vibration spectrum data.

[0037] Sound data is analyzed through power spectral density analysis to extract peak frequencies and harmonic components, and a fundamental frequency detection algorithm is used to identify the characteristic frequencies of equipment operation. After processing, vibration and sound data are synchronized and integrated into a multidimensional detection dataset using timestamps. This data is then stored in a time series database (such as InfluxDB) to support subsequent equipment status analysis and quality assessment.

[0038] In some embodiments, preprocessing the sound detection data and the vibration detection data to obtain the detection time domain signal includes:

[0039] The detected time domain signal is written into the time series database according to the storage format of the time series database.

[0040] Writing the detected time domain signals into the time series database can significantly improve the efficiency, reliability and scalability of the system.

[0041] In some embodiments, a batch insert method is used to package 1,000 detection time domain signal data points generated every 10 seconds into a single write request to reduce network overhead. For example, for a terminal machine's vibration signal, 1,000 data points are generated every 10 seconds (sampling rate 100Hz) and submitted to the time series database cluster at once through the batch write interface.

[0042] In some embodiments, the compression characteristics of the time series database are utilized to apply Delta encoding or Gorilla compression algorithm to continuous numerical data (such as vibration amplitude), so that the storage space is significantly reduced.

[0043] S103: Converting the detection time domain signal into a detection frequency domain signal through fast Fourier transform;

[0044] The pre-processed detection time domain signal is converted to the frequency domain using the Fast Fourier Transform (FFT) algorithm. To reduce spectrum leakage, the detection time domain signal is weighted with a Hanning window before conversion and normalized (divided by N / 2) to ensure amplitude accuracy. The resulting frequency domain spectrum is plotted with the horizontal axis representing frequency (0 to 500 kHz) and the vertical axis representing amplitude (unit: m / s 2 ).

[0045] Then, the detected frequency domain signal X[k] is squared to obtain the energy spectrum density: Px(f)=|X[k]| 2 ;

[0046] Then, the detection frequency domain signal is segmented and windowed (such as Hanning window) and averaged by Welch method to reduce noise interference;

[0047] After normalization, the resulting PSD curve is plotted with frequency on the horizontal axis and power (in dB / Hz) on the vertical axis. The PSD curve can be used to identify the primary frequency components and their energy distribution during equipment operation. For example, during the crimping process of a terminal machine, the motor's rotational frequency (e.g., 50 Hz) and its harmonics (100 Hz, 150 Hz) exhibit significant energy peaks in the PSD. Abnormal wear can also introduce high-frequency noise (>1 kHz), leading to a sharp increase in energy in the high-frequency band of the PSD.

[0048] In the frequency domain spectrogram, the main frequency components can be identified by peak detection algorithms (such as local maximum search):

[0049] The frequency point with the highest amplitude in the PSD is selected as the peak frequency. The peak frequency is usually related to the natural frequency of the key components of the equipment. For example, the peak frequency is 50Hz during normal operation. If an additional high-frequency peak (such as 2kHz) appears, it may indicate bearing wear or gear meshing abnormality.

[0050] Next, perform harmonic analysis. Specifically, check the amplitudes at integer multiples of the main frequency. If the amplitude of the double frequency (100Hz) of the 50Hz main frequency is abnormally high, it may indicate unstable equipment operation or the presence of nonlinear vibration sources.

[0051] In some embodiments, if multiple signals have high coherence at certain frequencies, this may indicate that the signal components at these frequencies are correlated. Analyzing the signal's phase information can provide additional information about the signal propagation path and device status. By calculating the signal's autocorrelation and cross-correlation functions, one can understand the signal's periodicity and correlation.

[0052] S104: extracting detection feature data reflecting the terminal machine operation status from the detection frequency domain signal;

[0053] In this embodiment, the detection characteristic data includes fundamental frequency, harmonics, abnormal frequency components, changes in power spectrum density, and noise level.

[0054] After obtaining the spectrum through a fast Fourier transform (FFT), the fundamental frequency—the frequency component with the highest amplitude in the spectrum—is first located. For example, the motor rotational frequency of a terminal machine (e.g., 50 Hz) is typically the fundamental frequency, and its amplitude reflects the stability of the equipment's primary drive components. Next, check for harmonics at integer multiples of the fundamental frequency (e.g., 100 Hz, 150 Hz). If the harmonic amplitude increases significantly (e.g., if the harmonic energy caused by motor electromagnetic noise exceeds 30% of the fundamental frequency), this may indicate problems with the equipment, such as electromagnetic interference, rotor eccentricity, or poor gear meshing. For example, if a terminal machine experiences increased rotor vibration due to bearing wear during operation, the amplitude of the double frequency (100 Hz) of the 50 Hz fundamental frequency in the spectrum may increase abnormally. Combined with power spectral density (PSD) analysis, it can be confirmed that the energy is concentrated in this frequency region, indicating the risk of bearing failure.

[0055] Then observe the sudden high-frequency components (such as >1kHz) and low-frequency fluctuations (such as <10Hz) in the spectrum. High-frequency spikes may be caused by internal impact, fracture or bearing defects in the equipment. For example, metal fatigue of the crimping die of a terminal machine may cause high-frequency impact vibration, which appears as a narrow-band peak in the range of 1-5kHz in the spectrum; while low-frequency fluctuations (such as fractional harmonics of the power frequency) may be related to loose mechanical structure or installation deviation. Coherence analysis is used to verify whether the abnormal frequency is highly correlated with the vibration signal. For example, a terminal machine has a 34Hz low-frequency fluctuation in the spectrum due to loose gearbox bolts, and the coherence between the vibration and sound signals at this frequency is >0.9, indicating that there is a resonance problem in the mechanical structure.

[0056] Then the PSD curve is used to observe the changes in energy distribution over time or working conditions. For example, when the terminal machine is operating normally, the PSD energy is mainly concentrated in the fundamental frequency and its harmonic region (such as 50Hz to 200Hz). However, when the equipment is abnormal (such as the crimping tool is worn), the energy in the high frequency band (>500Hz) increases sharply, indicating that friction or impact is intensified. By comparing the PSD data of different time periods (knowledge base [3]), the degree of energy distribution deviation can be quantified. For example, after a certain terminal machine has been running continuously for 24 hours, the energy density at 1kHz in the PSD curve increases from -40dB / Hz to -25dB / Hz, indicating that the tool wear has caused an increase in vibration noise and requires timely maintenance.

[0057] The noise level is then extracted. Specifically, the noise level is used to analyze the background noise level in the spectrum, that is, the average energy of non-target frequency components. If the background noise is too high (such as ambient noise > 60dB), it may mask key frequency characteristics. A-weighted filtering is used to simulate the human ear's auditory characteristics and evaluate the energy distribution of noise in the 1-8kHz range. If the noise energy in a specific frequency band (such as 2-4kHz) is found to be significantly higher than other areas, it may indicate that the device has high-frequency resonance or external interference sources (such as inverter electromagnetic noise). At this time, it is necessary to combine wavelet denoising or spatial filtering (adjusting the sensor position) to reduce the impact of noise and ensure the accuracy of spectrum analysis.

[0058] In some embodiments, extracting detection feature data reflecting the terminal machine operating state from the detection frequency domain signal includes:

[0059] Get the operating parameters of the terminal machine;

[0060] By analyzing the periodic changes of fundamental frequency and harmonics, the number of operating cycles of the terminal machine per unit time is calculated;

[0061] Use the number of operating cycles as a benchmark for actual production capacity;

[0062] The baseline value is corrected by operating parameters to generate the production capacity deviation rate.

[0063] By combining equipment operating parameters with real-time fundamental frequency cycle counts, a dynamic comparison between theoretical values ​​and actual operating data is achieved, avoiding statistical bias caused by sole reliance on sensor data and ensuring that production capacity benchmarks more closely reflect actual production conditions. The periodic variations in the fundamental frequency and harmonics directly reflect the regularity of the equipment's mechanical motion. By analyzing these fluctuations, it is possible to determine in real time whether the equipment is in a stable production state, thereby distinguishing between periods of effective production capacity and periods of idling or abnormal operation.

[0064] Specifically, obtaining the operating parameters of the terminal machine includes: real-time reading of the terminal machine PLC equipment operating parameters, such as the theoretical crimping cycle length (preset value 125ms / time); wire harness production count (the number of wire harnesses passed as fed back by the encoder); equipment start / stop status signal (running / idle / alarm), etc.

[0065] The cyclic stability of the terminal machine's operation is then analyzed by analyzing the amplitude variations of the fundamental frequency (e.g., 50 Hz) and its harmonics (100 Hz, 150 Hz, etc.) in the frequency domain signal. For example, if the amplitude fluctuation of the fundamental frequency under continuous operating conditions is less than 5%, it indicates stable motor speed. However, if the amplitude of a harmonic (e.g., the tripled frequency of 150 Hz) suddenly increases by 20%, it may indicate abnormal gear meshing or tool wear. Combined with timestamp alignment technology, the number of occurrences of the fundamental frequency within a unit of time (e.g., 1 minute) is counted to calculate the actual number of operating cycles (e.g., 60 occurrences of the 50 Hz fundamental frequency within 1 minute correspond to 60 complete cycles).

[0066] The number of operating cycles per unit time is then defined as the baseline value for actual production capacity. For example, if the theoretical design capacity of a terminal machine is 60 crimping operations per minute, and the actual measured number of cycles corresponding to the fundamental frequency is 58 times / minute, the baseline value is 58. This baseline value directly reflects the current processing efficiency of the equipment and needs to be further corrected based on the energy distribution of the harmonic components in the spectrum. If the harmonic energy ratio is abnormal (for example, the 150Hz harmonic energy exceeds 15% of the fundamental frequency), power spectral density (PSD) analysis is required to determine whether the cycle time is extended due to mechanical vibration, and thus adjust the baseline value.

[0067] In this embodiment, the baseline value is corrected by the actual operating parameters to generate the production capacity deviation rate. The specific steps are as follows:

[0068] Final capacity deviation rate = (calibrated baseline value - theoretical design value) / theoretical design value × 100%. For example, if the calibrated baseline value is 64.4 times / minute and the theoretical value is 60 times / minute, the deviation rate is +7.3%. If the deviation rate exceeds ±5% for three consecutive times, an early warning is triggered, indicating maintenance needs.

[0069] In some embodiments, after extracting the detection feature data reflecting the terminal machine operating state from the detection frequency domain signal, the method further includes:

[0070] Get the start and stop signals and operating parameters of the terminal machine;

[0071] By detecting the duration of the frequency region where energy is concentrated in the power spectrum density, the effective operation time period of the terminal machine is determined;

[0072] Distinguish effective working time and non-working time according to start and stop signals, operating parameters and effective operating time period;

[0073] Generate working time statistics report based on effective working time and non-working time.

[0074] This implementation integrates start / stop signals with power spectrum energy analysis to avoid misjudgments caused by relying solely on device on / off signals, ensuring the authenticity of working time statistics. Furthermore, by distinguishing between effective working time and non-working time, structured statistical reports are generated, reducing manual recording errors and meeting production audit requirements.

[0075] Specifically, the start and stop signals of the terminal machine are collected in real time through the inverter control terminals (such as DI1 / DI2), and the operating parameters are obtained in combination with sensors, including motor speed, crimping cycle time and load torque.

[0076] Next, based on power spectral density analysis, the duration of frequency regions where energy is concentrated in the spectrum is detected. The PSD energy concentration is then calculated using a sliding window method (e.g., a 10-second window). If a time period contains more than 80% of the fundamental frequency energy and lasts for more than 90% of the window period, it is considered to be an effective operating period. For example, during a two-hour continuous operation of a terminal machine, the PSD energy concentration in the 50Hz region lasted for 1.8 hours, indicating that the equipment's effective operating time accounted for 90%.

[0077] Then, the start-stop signal and PSD analysis results are combined to distinguish between effective working time and non-working time. For example:

[0078] The start / stop signal is "running" and the PSD energy is concentrated in the fundamental frequency area. The time period corresponding to the normal crimping action of the equipment is determined as the effective working time.

[0079] If the start / stop signal is "stop" or PSD energy dispersion (such as high-frequency noise >1kHz energy share >30%), the corresponding equipment is judged as non-working time due to no-load operation, fault shutdown or debugging stage.

[0080] Then, using timestamp alignment technology, we count the effective working time and non-working time throughout the day and generate a working time distribution chart. Based on the statistical results of effective working time and non-working time, we generate a working time statistics report.

[0081] S105: Matching the detection feature data with a preset fault feature model;

[0082] In this embodiment, the process of constructing the fault feature model includes:

[0083] Collect the sound data and vibration data of the terminal machine in the running state within a predetermined time to obtain the sound construction data and the vibration construction data;

[0084] Preprocessing the sound construction data and the vibration construction data to obtain a construction time domain signal;

[0085] Convert the constructed time domain signal into a constructed frequency domain signal through fast Fourier transform;

[0086] Extracting construction feature data from the construction frequency domain signal;

[0087] Construct a feature database of the terminal machine under normal operating conditions based on the constructed feature data;

[0088] Extract characteristic data of the terminal machine in a normal operating state from the characteristic database to obtain normal characteristic data;

[0089] Extract corresponding feature data from the feature database based on historical abnormal events to obtain abnormal feature data;

[0090] A machine learning algorithm is used to classify and train normal feature data and abnormal feature data to generate a fault feature model.

[0091] This embodiment extracts characteristic data of corresponding time periods in combination with historical abnormal events, and specifically labels fault types, so that the classification model can distinguish different fault modes with similar frequency domain features, thereby improving model specificity.

[0092] More specifically, the system collects sound and vibration data from the terminal machine during operation within a predetermined timeframe to generate sound and vibration data. Preprocessing the sound and vibration data to generate a constructed time-domain signal involves synchronously collecting sound and vibration data from the terminal machine during operation using a high-precision microphone and vibration sensor within a predetermined timeframe (e.g., one month of continuous operation). The preprocessing phase uses the 3σ criterion to eliminate outliers, wavelet threshold denoising to eliminate sensor noise, and time series interpolation to align asynchronous data, ultimately outputting a standardized constructed time-domain signal.

[0093] The constructed time domain signal is then converted into a constructed frequency domain signal through fast Fourier transform; extracting constructed feature data from the constructed frequency domain signal includes: converting the constructed time domain signal into a constructed frequency domain signal through fast Fourier transform (FFT), and extracting key features: fundamental frequency and harmonic characteristics, abnormal frequency components and power spectral density (PSD).

[0094] Then, based on the extracted frequency domain features, a feature database is constructed, which includes timestamp (millisecond level), device ID, sensor type (sound / vibration), and feature values ​​(fundamental frequency amplitude, PSD energy ratio, etc.).

[0095] Then, the data of the equipment during stable operation (such as 30 consecutive days of fault-free records) is screened from the feature database, and samples with a fundamental frequency energy ratio of >75% and a high-frequency noise ratio of <5% are extracted as the normal feature data set.

[0096] Based on historical fault events, we extract characteristic data from the corresponding time period as abnormal feature data. For example, in a bearing fault event, the energy proportion of the PSD high-frequency band (>5kHz) increased sharply from 3% to 18%, accompanied by a 98Hz envelope peak, marking it as an abnormal sample.

[0097] Then, a machine learning algorithm is used to classify and train the normal feature data and abnormal feature data to generate a fault feature model.

[0098] S106: If the detection feature data successfully matches the preset fault feature model, it is determined that the terminal machine has a fault and a fault warning signal is issued.

[0099] In some embodiments, further comprising:

[0100] Acquire the voiceprint signal of a standard quality product and perform time-frequency analysis on the voiceprint signal to obtain standard voiceprint feature data;

[0101] Construct a voiceprint matching model based on standard voiceprint feature data;

[0102] Acquire the real-time voiceprint signal of the product to be tested, and perform time-frequency analysis on the real-time voiceprint signal to obtain real-time voiceprint feature data;

[0103] Input the real-time voiceprint feature data into the voiceprint matching model for similarity comparison and calculate the matching score;

[0104] Determine whether the quality of the product to be tested is qualified based on the matching score and the preset threshold;

[0105] If the matching score exceeds the preset threshold, the quality of the product to be tested is determined to be qualified;

[0106] If the matching score does not exceed the preset threshold, the quality of the product to be tested is determined to be unqualified.

[0107] This embodiment upgrades quality inspection from "experience-based" to "data-driven" through time-frequency analysis and model matching of voiceprint signals, combining high precision, high speed and strong adaptability. It effectively solves the pain points of traditional quality inspection such as high missed detection rate, low efficiency and inconsistent standards, and significantly improves product consistency and yield.

[0108] The voiceprint matching model can use the support vector machine (SVM) model. Based on the similarity distribution output by the SVM model, a threshold of T = 0.85 is set. If the similarity ≥ T, it is judged as qualified; if it is < 0.85, a secondary test is triggered.

[0109] This embodiment uses sensors to collect time-domain signal data such as device vibration and temperature in real time. After preliminary preprocessing by the edge computing node, the raw data and preprocessed feature values ​​are batch-written into the time series database according to the time series database's row protocol format (e.g., device_metrics, device_id = A123vibration = 12.5, temperature = 45.21685432100000). Then, based on the stored historical data, a sliding window algorithm is used to calculate the real-time trend of the device's operating parameters. A statistical threshold method is used to detect abnormal fluctuations, and the analysis results are stored in the data storage unit of the relational database. The user interface displays the device status in real time through visual components (such as line charts and dashboards): the left panel dynamically plots the vibration amplitude curve in the form of a timeline, the right side displays key indicators such as the current temperature and operating time, and the list below pushes abnormal alarm information in real time. Users can retrieve historical data by time range or export analysis reports, achieving closed-loop management of the entire process from data collection and storage to visual display.

[0110] In some embodiments, a thermal imaging camera is used to capture the surface temperature distribution of the device in real time to obtain infrared thermal imaging data. Then, the temperature gradient features are extracted from the infrared thermal imaging data through local binary patterns (LBP), and the temperature anomaly index of the hot spot area (such as the deviation value from the baseline temperature) is calculated.

[0111] Then, a voiceprint encoder (CNN-LSTM network), vibration encoder (1D-CNN), and infrared encoder (2D-CNN) are constructed respectively to extract independent feature vectors of each modality (such as voiceprint feature dimension 128, vibration feature dimension 64, and infrared feature dimension 256).

[0112] Then, a multi-head cross attention module is used to calculate the similarity weight matrix between the voiceprint features and the vibration features. The calculation formula is:

[0113]

[0114] Among them, Q is the voiceprint feature query vector, K is the vibration feature key vector, and V is the vibration feature value vector; d k represents the dimension of the key vector; T represents the transpose operation of the vibration feature matrix K, ensuring that the dot product of Q and K is feasible.

[0115] Then, through the bidirectional InfoNCE loss, the feature spaces of voiceprint features and vibration features, voiceprint features and infrared features, and vibration features and infrared features are aligned to obtain multimodal features.

[0116] The aligned multimodal features are then input into the fully connected network (FCN), and the contribution of each modality is dynamically adjusted in combination with the gating mechanism (GatingNetwork).

[0117] Next, the multimodal features, processed by the fully connected network and gating mechanism, are fed into a classifier for identity recognition. During the training phase, the entire model is trained using a large amount of annotated multimodal data, optimizing its parameters so that the model can accurately determine the corresponding identity information based on the multimodal features. Furthermore, to improve the model's generalization capabilities, techniques such as data augmentation and regularization are employed. For data augmentation, voiceprint features are subjected to noise and speed-shifting, and vibration and infrared features undergo slight affine transformations. For regularization, an L2 regularization term is added to the loss function to prevent overfitting. During the testing phase, new multimodal data is fed into the trained model, which quickly and accurately outputs identity recognition results. Furthermore, to further validate the model's performance, experimental comparisons were conducted on multiple public datasets, and a comprehensive performance evaluation was conducted against existing single-modal and multimodal identity recognition models, including metrics such as precision, recall, and F1 score.

[0118] This embodiment, which uses a bidirectional InfoNCE loss to align multimodal features and incorporates a gating mechanism, demonstrates significant advantages in identity recognition tasks, achieving higher accuracy and greater robustness, enabling effective identification in complex environments. By combining a cross-modal attention mechanism with multimodal feature fusion and collaborative analysis of multi-source data, the robustness and generalization capabilities of industrial equipment fault detection are significantly improved.

[0119] Furthermore, an incremental learning framework is adopted to automatically add newly collected fault samples to the training set, and the model parameters are fine-tuned through the FTRL optimization algorithm.

[0120] Specifically, the system extracts nearly a week of historical data from a time series database as a basic training set, and uses stochastic gradient descent (SGD) to initialize model weights. Subsequently, incremental training is initiated every 100 newly annotated samples. The FTRL algorithm is used to balance the model's generalization ability for historical data with its adaptability to new samples. In particular, a time factor is introduced into the learning rate decay strategy to ensure that knowledge of early fault modes is not quickly forgotten. During training, the system automatically calculates the cumulative gradient values ​​of new and old samples and updates the L1 / L2 regularization parameters to mitigate the risk of overfitting. The optimized model parameters are synchronized to edge computing nodes, improving fault diagnosis accuracy in real time.

[0121] Using an incremental learning framework, newly collected fault samples are automatically added to the training set. The model parameters are fine-tuned using the FTRL optimization algorithm, enabling the model to quickly adapt to emerging fault modes. During the fine-tuning process, model performance metrics such as precision and recall on the validation set are continuously monitored to ensure stable and improved performance after fine-tuning. If a downward trend in model performance is detected during validation, the issue is immediately analyzed to determine whether the new samples contain noise or if the parameters of the FTRL optimization algorithm are improperly set. Appropriate measures are taken for different causes. If the cause is sample noise, the newly collected fault samples are cleaned and filtered. If the cause is an algorithm parameter issue, the FTRL algorithm parameters such as the learning rate and regularization coefficient are readjusted using methods such as grid search. After multiple rounds of fine-tuning and verification, the model reaches optimal performance, accurately identifying various new and existing faults, providing reliable technical support for subsequent fault warnings. Furthermore, a long-term model performance tracking mechanism is established to regularly collect new fault samples for incremental learning and continuously optimize the model to cope with ever-changing fault scenarios.

[0122] This embodiment can promptly detect potential equipment failures, reduce equipment downtime, and improve equipment availability and reliability by collecting and analyzing sound and vibration data in real time. By matching and analyzing voiceprint data with product quality, it is possible to detect product quality problems in real time, reduce defective rates, and improve production efficiency. By combining production capacity data statistics and working time statistics with fault warning functions, it is possible to optimize the operating efficiency of the equipment and improve the overall efficiency (OEE) of the equipment. By analyzing and storing long-term equipment operation data, it is possible to better manage the entire life cycle of the equipment and provide data support for equipment maintenance, updating, and optimization. At the same time, it reduces the workload of manual inspections and testing, and improves the automation and intelligence level of the production process.

[0123] See also Figure 2 This embodiment provides a terminal machine online analysis and early warning device 200, comprising:

[0124] The collecting unit 201 is used to collect the sound data and vibration data of the terminal machine during operation to obtain the sound detection data and vibration detection data;

[0125] A preprocessing unit 202 is used to preprocess the sound detection data and the vibration detection data to obtain a detection time domain signal;

[0126] A conversion unit 203 is configured to convert the detection time domain signal into a detection frequency domain signal through a fast Fourier transform;

[0127] The extraction unit 204 is used to extract detection feature data reflecting the operating status of the terminal machine from the detection frequency domain signal;

[0128] A matching unit 205 is configured to match the detection feature data with a preset fault feature model;

[0129] The determination unit 206 is configured to determine that a fault exists in the terminal machine and issue a fault warning signal if the detection feature data successfully matches a preset fault feature model.

[0130] Furthermore, the determining unit 206 includes:

[0131] The data acquisition subunit is used to collect the sound data and vibration data of the terminal machine in the running state within a predetermined time to obtain the sound construction data and the vibration construction data;

[0132] A first preprocessing subunit is used to preprocess the sound construction data and the vibration construction data to obtain a construction time domain signal;

[0133] A first conversion subunit, configured to convert the constructed time domain signal into a constructed frequency domain signal through a fast Fourier transform;

[0134] a feature extraction subunit, configured to extract construction feature data from the construction frequency domain signal;

[0135] A database construction subunit, configured to construct a feature database of the terminal machine under normal operating conditions based on the constructed feature data;

[0136] A data extraction subunit is used to extract characteristic data of the terminal machine in a normal operating state from the characteristic database to obtain normal characteristic data;

[0137] an abnormal feature data acquisition subunit, configured to extract corresponding feature data from the feature database based on historical abnormal events to obtain abnormal feature data;

[0138] The training subunit is used to use a machine learning algorithm to perform classification training on the normal feature data and the abnormal feature data to generate a fault feature model.

[0139] Furthermore, the detection characteristic data includes fundamental frequency, harmonics, abnormal frequency components, changes in power spectrum density and noise level.

[0140] Furthermore, the extraction unit 204 includes:

[0141] Parameter acquisition subunit, used to obtain the operating parameters of the terminal machine;

[0142] The analysis subunit is used to calculate the number of operating cycles of the terminal machine per unit time by analyzing the periodic changes of the fundamental frequency and harmonics;

[0143] A reference value acquisition subunit, configured to use the number of operating cycles as a reference value of actual production capacity;

[0144] The correction subunit is used to correct the reference value according to the operating parameter to generate a production capacity deviation rate.

[0145] Furthermore, the extraction unit 204 includes:

[0146] The operating parameter acquisition subunit is used to obtain the start and stop signals and operating parameters of the terminal machine;

[0147] a determination subunit, configured to determine an effective operation time period of the terminal machine by detecting a duration of a frequency region in which energy is concentrated in the power spectrum density;

[0148] A distinguishing subunit, configured to distinguish effective working time and non-working time according to the start / stop signal, the operating parameters and the effective operating time period;

[0149] The report generating subunit is used to generate a working time statistics report based on the effective working time and non-working time.

[0150] Furthermore, it also includes:

[0151] A time-frequency analysis unit, configured to obtain a voiceprint signal of a standard quality product and perform time-frequency analysis on the voiceprint signal to obtain standard voiceprint feature data;

[0152] A model building unit, configured to build a voiceprint matching model based on the standard voiceprint feature data;

[0153] A signal acquisition unit is used to acquire a real-time voiceprint signal of the product to be detected, and perform time-frequency analysis on the real-time voiceprint signal to obtain real-time voiceprint feature data;

[0154] a comparison unit, configured to input the real-time voiceprint feature data into the voiceprint matching model for similarity comparison and calculate a matching score;

[0155] A quality judgment unit is used to judge whether the quality of the product to be tested is qualified based on the matching score and the preset threshold;

[0156] A qualified judgment unit, used to judge that the quality of the product to be tested is qualified if the matching score exceeds a preset threshold;

[0157] The unqualified judgment unit is used to judge that the quality of the product to be tested is unqualified if the matching score does not exceed a preset threshold.

[0158] Furthermore, the pre-processing unit 202 includes:

[0159] The writing subunit is used to write the detected time domain signal into the time series database according to the storage format of the time series database.

[0160] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-mentioned devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0161] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, can implement the methods provided in the above embodiments. The storage medium may include: a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, among other media capable of storing program code.

[0162] The present invention further provides a computer device that may include a memory and a processor. The memory stores a computer program, and the processor, when invoking the computer program in the memory, can implement the method provided in the above embodiment. Of course, the computer device may also include various network interfaces, a power supply, and other components.

[0163] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0164] It should also be noted that, in this specification, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprising" or any other variations thereof are intended to cover non-exclusive.

[0165] Inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a..." does not preclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

Claims

1. A terminal machine online analysis and early warning method, characterized in that: include: Collect the sound data and vibration data of the terminal machine during operation to obtain sound detection data and vibration detection data; Preprocessing the sound detection data and the vibration detection data to obtain a detection time domain signal; Converting the detection time domain signal into a detection frequency domain signal through fast Fourier transform; Extracting detection feature data reflecting the terminal machine operating state from the detection frequency domain signal; Matching the detection feature data with a preset fault feature model; If the detection feature data successfully matches the preset fault feature model, it is determined that the terminal machine has a fault and a fault warning signal is issued.

2. The terminal machine online analysis and early warning method according to claim 1 is characterized in that: The process of constructing the fault feature model includes: Collect the sound data and vibration data of the terminal machine in the running state within a predetermined time to obtain the sound construction data and the vibration construction data; Preprocessing the sound construction data and the vibration construction data to obtain a construction time domain signal; Convert the constructed time domain signal into a constructed frequency domain signal through fast Fourier transform; Extracting construction feature data from the construction frequency domain signal; Constructing a feature database of the terminal machine under normal operating conditions based on the constructed feature data; Extracting characteristic data of the terminal machine in a normal operating state from the characteristic database to obtain normal characteristic data; Extracting corresponding feature data from the feature database based on historical abnormal events to obtain abnormal feature data; A machine learning algorithm is used to perform classification training on the normal feature data and the abnormal feature data to generate a fault feature model.

3. The terminal machine online analysis and early warning method according to claim 1 is characterized in that: The detection characteristic data includes fundamental frequency, harmonics, abnormal frequency components, changes in power spectrum density and noise level.

4. The terminal machine online analysis and early warning method according to claim 3 is characterized in that: After extracting the detection feature data reflecting the terminal machine operation state from the detection frequency domain signal, the method includes: Get the operating parameters of the terminal machine; By analyzing the periodic changes of fundamental frequency and harmonics, the number of operating cycles of the terminal machine per unit time is calculated; Taking the number of operating cycles as a benchmark value for actual production capacity; The reference value is corrected by the operating parameter to generate a production capacity deviation rate.

5. The terminal machine online analysis and early warning method according to claim 3 is characterized in that: After extracting the detection feature data reflecting the terminal machine operation state from the detection frequency domain signal, the method further includes: Get the start and stop signals and operating parameters of the terminal machine; By detecting the duration of the frequency region where energy is concentrated in the power spectrum density, the effective operation time period of the terminal machine is determined; Distinguishing effective working time and non-working time according to the start / stop signal, the operating parameters and the effective operating time period; Generate a working time statistics report based on the effective working time and non-working time.

6. The terminal machine online analysis and early warning method according to claim 1 is characterized in that: Also includes: Acquire a voiceprint signal of a standard quality product, and perform time-frequency analysis on the voiceprint signal to obtain standard voiceprint feature data; Constructing a voiceprint matching model based on the standard voiceprint feature data; Acquire a real-time voiceprint signal of the product to be tested, and perform time-frequency analysis on the real-time voiceprint signal to obtain real-time voiceprint feature data; Inputting the real-time voiceprint feature data into the voiceprint matching model for similarity comparison and calculating the matching score; Determine whether the quality of the product to be tested is qualified based on the matching score and the preset threshold; If the matching score exceeds the preset threshold, the quality of the product to be tested is determined to be qualified; If the matching score does not exceed the preset threshold, the quality of the product to be tested is determined to be unqualified.

7. The terminal machine online analysis and early warning method according to claim 1 is characterized in that: The preprocessing of the sound detection data and the vibration detection data to obtain the detection time domain signal includes: The detected time domain signal is written into the time series database according to the storage format of the time series database.

8. A terminal machine online analysis and early warning device, characterized in that: include: The acquisition unit is used to collect the sound data and vibration data of the terminal machine during operation to obtain the sound detection data and vibration detection data; A preprocessing unit, configured to preprocess the sound detection data and the vibration detection data to obtain a detection time domain signal; A conversion unit, configured to convert the detection time domain signal into a detection frequency domain signal through a fast Fourier transform; An extraction unit, configured to extract detection feature data reflecting the terminal machine's operating status from the detection frequency domain signal; A matching unit, configured to match the detection feature data with a preset fault feature model; The determination unit is configured to determine that a fault exists in the terminal machine and issue a fault warning signal if the detection feature data successfully matches a preset fault feature model.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the terminal machine online analysis and early warning method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, causes the processor to execute the terminal machine online analysis and early warning method according to any one of claims 1 to 7.

Images