A method and system for diagnosing production defects of a quartz crystal resonator

Abstract

This invention relates to the field of defect diagnosis technology, and more particularly to a method and system for diagnosing manufacturing defects in quartz crystal resonators. The method includes the following steps: extracting micro-acoustic response signals from each processing stage using an ultrasonic excitation and response monitoring device; performing acoustic time-frequency structure analysis and response trajectory trend analysis and extraction to obtain a multi-stage crystal acoustic waveform behavior trajectory sequence; deploying a multi-angle structured light interferometer array for simultaneous acoustic monitoring and visual acquisition; calculating optical waveform morphology response deviations and extracting optical waveform morphology response deviation segments; performing acoustic-optical behavior frequency response analysis on the multi-stage crystal acoustic waveform behavior trajectory sequence and optical waveform morphology response deviation segments; fitting acoustic-optical mismatch indices; and constructing an acoustic-optical fusion feature deviation database. This invention improves the production yield of quartz crystal resonators and reduces production line waste and energy consumption through accurate and efficient defect diagnosis and localization.

Description

Technical Field

[0001] This invention relates to the field of defect diagnosis technology, and in particular to a method and system for diagnosing manufacturing defects in quartz crystal resonators. Background Technology

[0002] Quartz crystal resonators, as key components for high-precision frequency control and time reference, are widely used in communications, navigation, and smart devices. During their manufacturing process, the processing precision and quality of the crystal directly affect the resonator's performance stability and reliability. However, due to the complex manufacturing process of quartz crystal resonators, involving multiple precision machining steps such as cutting, polishing, and etching, even minor process fluctuations or equipment malfunctions can easily lead to crystal morphology defects, stress concentration, and even structural damage. These defects not only reduce the resonator's quality factor and frequency stability but may also shorten product lifespan and cause performance drift, severely impacting the efficiency and stability of end-user equipment.

[0003] Currently, defect diagnosis in quartz crystal resonators mainly relies on manual inspection, traditional optical inspection equipment, and some online acoustic monitoring methods. While these methods can detect obvious defects to a certain extent, they often suffer from insufficient detection sensitivity, slow response speed, and inaccurate defect localization, making it difficult to meet the demands of high-throughput, real-time online inspection. Furthermore, traditional defect diagnosis technologies often depend on single-sensor information, neglecting the multi-physical coupling characteristics during crystal processing. This results in weak identification capabilities for hidden and complex defects, failing to comprehensively reflect the actual state of the production line. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method and system for diagnosing manufacturing defects in quartz crystal resonators, thereby resolving at least one of the aforementioned technical problems.

[0005] To achieve the above objectives, the present invention provides a method for diagnosing manufacturing defects in quartz crystal resonators, comprising the following steps:

[0006] Step S1: Extract the micro-acoustic response signals of each processing stage based on the ultrasonic excitation and response monitoring device, perform acoustic time-frequency structure analysis and response trajectory trend analysis and extraction to obtain a multi-stage crystal acoustic behavior trajectory sequence;

[0007] Step S2: Deploy a multi-angle structured light interferometer array to perform acoustic monitoring and synchronous visual acquisition, calculate the light ripple morphology response deviation, and extract the light ripple morphology response deviation segment;

[0008] Step S3: Perform acoustic-optical behavior frequency response analysis on the multi-stage crystal acoustic pattern behavior trajectory sequence and optical pattern morphology response deviation segment, and fit the acoustic-optical mismatch index to construct an acoustic-optical fusion feature deviation database.

[0009] Step S4: Based on the acoustic-optical fusion feature deviation database, locate the three-dimensional defect points and perform joint analysis of acoustic reflection morphology and optical interference morphology to obtain the defect point type characteristics;

[0010] Step S5: Perform time series analysis on the defect point type characteristics and analyze the evolution of the time series deviation to generate the structural deviation trend curve and the expected electrical performance index drift value.

[0011] Step S6: Based on the structural deviation trend curve and the expected electrical performance index drift value, perform performance impact diagnosis and iterative closed-loop control optimization of the production line to execute production defect diagnosis.

[0012] This specification provides a manufacturing defect diagnosis system for quartz crystal resonators, used to perform the manufacturing defect diagnosis method for quartz crystal resonators as described above, including:

[0013] The acoustic analysis module is used to extract the micro-acoustic response signals of each processing stage based on the ultrasonic excitation and response monitoring device, perform acoustic time-frequency structure analysis and response trajectory trend analysis and extraction, so as to obtain a multi-stage crystal acoustic pattern behavior trajectory sequence.

[0014] The optical morphology response module is used to deploy a multi-angle structured light interferometer array for acoustic monitoring and synchronous visual acquisition, calculate the optical morphology response deviation, and extract the optical morphology response deviation segment.

[0015] The acoustic-optic mismatch module is used to perform acoustic-optic behavior frequency response analysis on the multi-stage crystal acoustic pattern behavior trajectory sequence and optical pattern morphology response deviation segment, and to fit the acoustic-optic mismatch index to construct an acoustic-optic fusion feature deviation database.

[0016] The defect point localization module is used to locate three-dimensional defect points based on the acoustic-optical fusion feature deviation database, and to perform joint analysis of acoustic reflection morphology and optical interference morphology to obtain defect point type characteristics.

[0017] The deviation situation evolution module is used to perform time-series deviation analysis based on the characteristics of defect point type, and to perform time-series deviation situation evolution analysis to generate structural deviation trend curves and expected electrical performance index drift values.

[0018] The performance diagnostic module is used to diagnose the performance impact based on the structural deviation trend curve and the expected electrical performance index drift value, and to perform iterative closed-loop control optimization of the production line to execute production defect diagnosis operations.

[0019] The specific benefits of this invention are as follows: High-resolution time-frequency analysis of the acoustic behavior of quartz crystals during multiple key processing stages such as cutting, grinding, and coating enables real-time capture of changes in their inherent acoustic response; by extracting the "acoustic behavior trajectory sequence," an acoustic "behavioral fingerprint" with crystal characteristics can be constructed, establishing an accurate behavioral baseline for subsequent anomaly identification; a complete "multi-stage acoustic evolution map" is formed during processing, achieving a technological breakthrough in monitoring the entire process from early stages to finished product. Real-time reconstruction of the three-dimensional micro-morphology of the crystal surface and subsurface can be achieved through multi-angle structured light interferometry array acquisition, significantly improving spatial resolution and anomaly identification rate; combined with an acoustic synchronous calibration mechanism, the complete correspondence between optical acquisition and acoustic behavior is ensured, improving the accuracy of cross-modal analysis; extraction of "optical deviation segments" can capture surface morphology perturbations caused by uneven processing pressure and tool abnormalities in advance, providing data support for early defect identification. A joint mapping relationship between acoustic behavior characteristics and optical morphological response was established, enabling the construction of a cross-modal fusion diagnostic model. Through fitting of acousto-optic mismatch indices, latent abnormal behaviors, such as submicron-level crack initiation or poor stress coupling, which are difficult to identify from a single mode alone, can be identified. The construction of an acousto-optic fusion deviation database facilitates the automatic classification and generalization of defect patterns by artificial intelligence algorithms, enhancing the self-learning ability and reliability of the monitoring system. Through spatial reconstruction and inversion after multimodal information fusion, precise three-dimensional positioning of defect points (XYZ coordinate level) can be achieved with errors below the micrometer level. Acoustic reflection behavior reveals the internal structure of defects, while optical interferometry reveals surface stress or morphological anomalies. Joint analysis enables high-precision defect classification (such as cracks, foreign objects, bubbles, stress zones, etc.). The output "defect type characteristics" support high-precision source tracing analysis and fault location, helping to quickly pinpoint problematic processes and abnormal process parameters. By performing time-series modeling of the defect evolution trajectory during processing, it is possible to identify whether a defect is caused by a single sudden occurrence or by process accumulation. A "structural deviation trend curve" is constructed to quantitatively predict the future impact of the current defect state on key performance characteristics such as crystal thickness consistency, frequency stability, and Q value. The output "electrical performance drift value" enables proactive performance prediction, providing quantitative indicators for process optimization and early maintenance, reducing experimental costs and rework costs. This achieves closed-loop control throughout the entire process of "anomaly identification → early warning analysis → process adjustment." The system can push control commands (such as tool pressure reduction, vibration compensation, and cooling rhythm adjustment) based on deviation trends, forming a real-time feedback mechanism. This substantially improves the production yield and consistency of quartz crystal resonators, effectively shortening the quality control cycle and reducing waste and energy consumption. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of the steps in the method for diagnosing manufacturing defects in a quartz crystal resonator according to the present invention.

[0021] Figure 2This is a detailed flowchart illustrating the implementation steps of step S1.

[0022] Figure 3 This is a detailed flowchart illustrating the implementation steps of step S2;

[0023] Figure 4 This is a flowchart illustrating the detailed implementation steps of step S3. Detailed Implementation

[0024] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0025] This application provides a method and system for diagnosing manufacturing defects in quartz crystal resonators. The execution entities of the method and system for diagnosing manufacturing defects in quartz crystal resonators include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, and network upload devices that can be considered general computing nodes in this application. The data processing platform includes, but is not limited to, at least one of an audio-visual management system, an information management system, and a cloud-based data management system.

[0026] Please see Figures 1 to 4 This invention provides a method for diagnosing manufacturing defects in quartz crystal resonators, comprising the following steps:

[0027] Step S1: Extract the micro-acoustic response signals of each processing stage based on the ultrasonic excitation and response monitoring device, perform acoustic time-frequency structure analysis and response trajectory trend analysis and extraction to obtain a multi-stage crystal acoustic behavior trajectory sequence;

[0028] Step S2: Deploy a multi-angle structured light interferometer array to perform acoustic monitoring and synchronous visual acquisition, calculate the light ripple morphology response deviation, and extract the light ripple morphology response deviation segment;

[0029] Step S3: Perform acoustic-optical behavior frequency response analysis on the multi-stage crystal acoustic pattern behavior trajectory sequence and optical pattern morphology response deviation segment, and fit the acoustic-optical mismatch index to construct an acoustic-optical fusion feature deviation database.

[0030] Step S4: Based on the acoustic-optical fusion feature deviation database, locate the three-dimensional defect points and perform joint analysis of acoustic reflection morphology and optical interference morphology to obtain the defect point type characteristics;

[0031] Step S5: Perform time series analysis on the defect point type characteristics and analyze the evolution of the time series deviation to generate the structural deviation trend curve and the expected electrical performance index drift value.

[0032] Step S6: Based on the structural deviation trend curve and the expected electrical performance index drift value, perform performance impact diagnosis and iterative closed-loop control optimization of the production line to execute production defect diagnosis.

[0033] In the embodiments of the present invention, see Figure 1 This is a flowchart illustrating the steps of a method for diagnosing manufacturing defects in a quartz crystal resonator according to the present invention. In this example, the steps of the method for diagnosing manufacturing defects in a quartz crystal resonator include:

[0034] Step S1: Extract the micro-acoustic response signals of each processing stage based on the ultrasonic excitation and response monitoring device, perform acoustic time-frequency structure analysis and response trajectory trend analysis and extraction to obtain a multi-stage crystal acoustic behavior trajectory sequence;

[0035] In this embodiment, piezoelectric ultrasonic exciters and high-sensitivity acoustic sensors are deployed at key workstations in the quartz crystal resonator production line, including cutting, grinding, and coating. The exciters emit ultrasonic signals with a frequency range of 1-10MHz, and the excitation power is controlled between 0.1-1W to ensure no damage to the crystal structure. The acoustic sensors are broadband piezoelectric ceramic sensors with a frequency response range of 0.1-20MHz, a sensitivity of -40dB, and a sampling rate of 50MHz, capable of capturing minute acoustic response changes within the crystal. Time-frequency analysis is performed on the acquired acoustic signals using a combination of short-time Fourier transform and wavelet transform to extract frequency domain features, including the dominant frequency component, harmonic components, and spectral centroid. Empirical mode decomposition (EMD) is used to decompose the complex acoustic signals into multiple intrinsic mode functions (EMFs), each representing a different physical mechanism. A sliding window technique is used to segment the time-series signal, with a window length of 10ms and an overlap rate of 50%, ensuring a balance between time and frequency resolution. By establishing an acoustic feature vector sequence, including parameters such as frequency center fc, bandwidth B, peak amplitude A, and phase φ, a acoustic signature template for each processing stage is constructed, ultimately resulting in a multi-stage crystal acoustic signature behavior trajectory sequence containing timestamps, frequency features, and amplitude features.

[0036] Step S2: Deploy a multi-angle structured light interferometer array to perform acoustic monitoring and synchronous visual acquisition, calculate the light ripple morphology response deviation, and extract the light ripple morphology response deviation segment;

[0037] In this embodiment, a multi-angle monitoring array consisting of 3-5 high-precision structured light interferometers is deployed, synchronized with the acoustic monitoring system. Each interferometer is equipped with a He-Ne laser (wavelength 632.8nm, power 2mW), a beam splitter, a reference mirror, and a high-resolution CCD camera (resolution 1024×1024 pixels, pixel size 4.65μm). The working distance of the interferometer is set to 50-100mm, the numerical aperture is 0.25, and the theoretical resolution can reach 0.5μm. Phase-shifting interferometry is used, with a λ / 4 phase shift controlled by a piezoelectric ceramic actuator to obtain four phase-shifted interferograms, eliminating the influence of background light and system errors. A theoretical optical field model based on the material properties of quartz crystal is established, considering the birefringence and elasto-optical effects of the crystal, to calculate the ideal interference fringe distribution. In actual measurements, the contrast of the interference fringes is usually between 0.3 and 0.8. The phase information of the fringes is extracted using image processing algorithms to calculate the surface height change. The least squares method is used to match the actual interferometric image with the theoretical template, calculate the residual distribution map, and identify anomalous regions where the deviation exceeds a threshold (usually set to λ / 10, i.e., 63nm). Through morphological processing and connected component analysis, geometric feature parameters of the anomalous regions are extracted, including area, perimeter, aspect ratio, centroid coordinates, etc., forming a dataset of optical ripple morphological response deviation segments.

[0038] Step S3: Perform acoustic-optical behavior frequency response analysis on the multi-stage crystal acoustic pattern behavior trajectory sequence and optical pattern morphology response deviation segment, and fit the acoustic-optical mismatch index to construct an acoustic-optical fusion feature deviation database.

[0039] In this embodiment, an acoustic-optical synchronization analysis framework is established to ensure that the time alignment accuracy of the two modal data is at the microsecond level. Cross-correlation analysis is used to calculate the correlation between the acoustic signal and the optical deviation signal, with a correlation coefficient threshold of 0.7 or higher considered as a significant correlation. Frequency domain analysis reveals that the dominant frequency component of the acoustic signal is typically in the range of 2-8 MHz, while the frequency variation of the optical deviation signal is relatively low, in the range of 0.1-1 kHz. Wavelet coherence analysis is used to study the coherence of the acoustic-optical signals at different frequency scales, identifying key frequency coupling intervals. An acoustic-optical mismatch index system is established, including key parameters such as time offset Δt (typical value 0.1-5 ms), spectral center shift Δfc (typical value 50-500 Hz), amplitude ratio deviation ΔA / A (typical value 5%-30%), and phase difference offset Δφ (typical value π / 8-π / 2). Multiple linear regression and support vector machine methods are used to fit and model the acoustic-optical mismatch index, establishing a mapping relationship from acoustic feature vectors to optical deviation. The coefficient of determination R of the regression model is... 2The accuracy reached above 0.85. By establishing a multi-dimensional data structure that includes time series, acoustic features, optical features, and mismatch indicators, an acoustic-optical fusion feature deviation database was constructed to provide data support for subsequent defect localization.

[0040] Step S4: Based on the acoustic-optical fusion feature deviation database, locate the three-dimensional defect points and perform joint analysis of acoustic reflection morphology and optical interference morphology to obtain the defect point type characteristics;

[0041] In this embodiment, multimodal information from the acoustic-optical fusion feature deviation database is used to determine the spatial coordinates of the defect point using a three-dimensional reconstruction algorithm. Acoustic localization employs the Time Difference of Arrival (TDOA) method, based on the propagation speed of ultrasound in quartz crystal (longitudinal wave velocity 5760 m / s, transverse wave velocity 3760 m / s). The depth information of the defect point is calculated by the time difference of reflected signals received by multiple sensors, achieving a localization accuracy of 10 μm. Optical localization uses multi-angle measurements from a structured light interferometer array, utilizing the principle of triangulation to determine the XY coordinates of the defect point, achieving a lateral resolution of 1 μm. Combining acoustic depth information and optical lateral position information, a three-dimensional spatial coordinate system for the defect point is constructed. For defect type identification, an acoustic reflection morphology analysis model is established. The nature of the defect is determined based on the amplitude, phase, and spectral characteristics of the reflected signal. Microcracks exhibit high-frequency scattering characteristics, bubble defects exhibit low-frequency reflection characteristics, and stress concentration exhibits spectral broadening characteristics. Optical interferometry morphology analysis identifies defect types through surface height variation patterns. Point defects exhibit local height abrupt changes, linear defects exhibit continuous height gradient changes, and planar defects exhibit large-area height deviations. Establish a defect feature vector, including spatial coordinates (X, Y, Z), acoustic features (reflection coefficient, spectral features), optical features (height deviation, shape features), and comprehensive confidence level, to form a defect point type feature database.

[0042] Step S5: Perform time series analysis on the defect point type characteristics and analyze the evolution of the time series deviation to generate the structural deviation trend curve and the expected electrical performance index drift value.

[0043] In this embodiment, a time-series analysis model of defect evolution is established based on temporal information from the defect type feature database. Time series analysis methods, including the Autoregressive Moving Average (ARIMA) model and Long Short-Term Memory (LSTM) network, are used to model the changes in the number, location, and type of defects over time. A sliding time window (window size set to 1 / 10 of the processing cycle, typically 30 seconds) is used to segment the defect evolution process, identifying key time points and evolution rates of defect generation. A defect cumulative effect model is established, considering the influence weights of different types of defects on the performance of the quartz crystal resonator: microcracks have a weight of 0.8 on the Q value, bubble defects have a weight of 0.6 on frequency stability, and stress concentration has a weight of 0.7 on the temperature coefficient. Using Monte Carlo simulation, the electrical performance indicators of the final product are predicted based on the statistical characteristics of defect distribution, including resonant frequency drift (typical prediction range ±10ppm), Q value change (typical prediction range ±5%), and temperature coefficient deviation (typical prediction range ±2ppm / ℃). The structural deviation trend curve is generated using exponential smoothing and trend decomposition methods. The curve comprises three components: linear trend, seasonal cycle, and random fluctuation, achieving a trend prediction accuracy of over 90%. The final output includes a structural deviation trend curve with a time axis, deviation magnitude, confidence interval, and the expected range of electrical performance index drift values.

[0044] Step S6: Based on the structural deviation trend curve and the expected electrical performance index drift value, perform performance impact diagnosis and iterative closed-loop control optimization of the production line to execute production defect diagnosis.

[0045] In this embodiment, a performance impact diagnostic model is established based on the structural deviation trend curve and the expected electrical performance index drift value. Through comparative analysis with product specifications, alarm thresholds for performance indicators are set: the frequency accuracy alarm threshold is set to ±5ppm, the Q value alarm threshold is set to 90% of the design value, and the temperature coefficient alarm threshold is set to ±1ppm / ℃. A risk assessment matrix method is used to classify product quality risks according to the degree of defect impact and probability of occurrence. High-risk products require rework or scrapping, medium-risk products require enhanced monitoring of subsequent processes, and low-risk products can proceed normally. An iterative closed-loop control system for the production line is established, including four stages: real-time monitoring, data analysis, decision support, and control execution. When a deviation trend is detected to exceed a preset threshold, the system automatically triggers process parameter adjustments, such as reducing the cutting speed by 10%-20%, adjusting the grinding pressure by ±0.1MPa, and optimizing the coating temperature by ±5℃. Through an interface with the Manufacturing Execution System (MES), automatic adjustment of production parameters and dynamic optimization of the production plan are achieved. A historical data learning mechanism is established to continuously optimize the predictive model parameters using past production data, improving diagnostic accuracy and control effectiveness. The system also features an alarm function. When the defect rate exceeds a set threshold (e.g., 5%), it automatically sends an alarm message to quality management personnel and records a detailed diagnostic report. Through continuous closed-loop control optimization, the production line achieves adaptive adjustment, ensuring the stability and consistency of quartz crystal resonator product quality, and ultimately completing the entire process of production defect diagnosis.

[0046] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:

[0047] Identify the processing flow of the quartz crystal resonator production line and mark the key processing equipment;

[0048] The key processing equipment is monitored for acoustic vibration in real time using an ultrasonic excitation and response monitoring device, and the micro-acoustic response signals of each processing stage are extracted.

[0049] The micro-acoustic response signal is subjected to environmental noise adaptive filtering to generate a filtered and optimized response signal;

[0050] Acoustic time-frequency structure analysis is performed on the filtered optimized response signal to generate time-frequency response characteristics for multiple processing stages;

[0051] Obtain equipment processing instructions, divide time segments based on the equipment processing instructions, and generate a crystal processing stage division map;

[0052] Based on the crystal processing stage division map, the acoustic behavior response trajectory trend analysis of the time-frequency response characteristics is performed and extracted to obtain a multi-stage crystal acoustic waveform trajectory sequence.

[0053] In this embodiment, the entire production process of quartz crystal resonators is systematically identified, including stages such as raw material processing of quartz crystal rods, crystal cutting, rough grinding, fine grinding, polishing, electrode evaporation, frequency fine-tuning, cleaning, packaging, and aging testing. Each processing stage corresponds to specific equipment and its operating parameters, which directly affect the frequency stability, Q value, and resonance characteristics of the finished crystal. Therefore, it is necessary to identify the equipment nodes throughout the entire process through on-site surveys, equipment layout analysis, and process records in the MES system, and to identify the key equipment with the greatest impact on product quality by combining yield data. During the identification process, special attention should be paid to equipment such as crystal cutting wire saws (speed approximately 3000 rpm), double-sided grinding machines (loading force up to 500N), fine polishing machines (speed between 20-50 rpm), vacuum evaporation systems (target current controlled at 300-500mA), and laser fine-tuning instruments. Through process sensitivity analysis, fault history, and process parameter fluctuations of these devices, 8-10 pieces of equipment that are most critical in terms of structure formation and frequency determination are finally identified. These key devices are the primary targets for subsequent acoustic monitoring, process behavior modeling, and defect prediction. Device numbers and process sequences are mapped into the monitoring system database, serving as the logical entry point for real-time monitoring deployment. High-frequency ultrasonic excitation and response monitoring devices are deployed to collect weak acoustic vibration signals during equipment operation, characterizing its mechanical state and process behavior. Piezoelectric ceramic transducers with a center frequency of 1MHz are used as ultrasonic transmitting and receiving elements. The excitation signal is generated as a 10Vpp pulse wave with a period of 100μs by a function generator, and the pulse repetition frequency is set to 20Hz to minimize interference to the equipment itself. The received signal is processed by a preamplifier with a gain of 40dB and then input to a high sampling rate (500kHz) data acquisition card (such as NIPXIe-4464), achieving alignment with the equipment's operating timestamp through a synchronous acquisition system. Transducers are fixed to the mechanical frame, rotating bearings, or bases of each device, and coupling adhesive or magnetic bases are used to ensure signal stability and coupling efficiency. The data acquisition cycle is set according to the process cycle. For example, in the rough grinding stage, it is set to acquire data once every 60 seconds, with each acquisition lasting 3 seconds; in the polishing stage, the acquisition cycle is increased to once every 30 seconds to capture subtle changes in state. The raw signals extracted in this way contain nonlinear features such as structural shock waves, micro-friction sounds, and superimposed vibration waves, which are the core carriers reflecting process changes and potential defects.

[0054] Because complex background noise sources exist in actual production workshops, such as fan noise, motor whistling, and human operation sounds, these interference signals can significantly mask the micro-sound response generated by the target equipment. Therefore, it is necessary to perform adaptive environmental noise filtering on the collected raw signals. This step uses an LMS (Least Mean Squares) adaptive filter combined with Independent Component Analysis (ICA) technology for multi-channel noise reduction. First, background noise samples are collected under no-load conditions in the workshop to establish a reference noise signal. Then, during the operation of the target equipment, this reference noise is used as input and input together with the target signal into the LMS filtering network. The filter weights are dynamically adjusted to effectively suppress the reference noise component. ICA is used to separate the independent sound source components of the equipment itself from the mixed signal, further eliminating interference components generated by non-target equipment. In experimental verification, the signal-to-noise ratio of the original signal was approximately 12-15 dB, which was improved to over 20-25 dB after processing. Before and after filtering, the quality of the signal was compared by calculating indicators such as the root mean square value, kurtosis, and spectral entropy in the time domain. The filtering fidelity was determined through expert evaluation to ensure the preservation of the device's inherent characteristics, providing a high-quality input signal foundation for subsequent acoustic time-frequency analysis. Time-frequency structure analysis was used to process the signal and extract multi-dimensional acoustic feature vectors. Short-time Fourier transform (STFT) was used to analyze the energy distribution of the signal, with a window length of 512 points and an overlap rate of 75%, yielding an energy evolution map of the signal in the frequency and time dimensions. Subsequently, continuous wavelet transform (CWT) was introduced for multi-resolution analysis, using the Morlet wavelet kernel function to analyze the variation trend in the frequency band from 1-50kHz to obtain fine local features. To further enhance the signal's ability to identify nonlinear behavior, empirical mode decomposition (EMD) was used to decompose the signal into multiple intrinsic mode functions (IMFs), extracting modes containing device-specific behavioral characteristics. For example, IMF2 and IMF3 typically contain high-frequency local friction vibration information. Within each processing stage, the time-domain statistical characteristics (such as mean, variance, skewness, kurtosis), frequency-domain characteristics (such as dominant frequency, spectral centroid, spectral entropy), and time-frequency characteristics (such as local time-frequency energy, wave packet entropy, etc.) of the signal are calculated respectively, and a total of 48-64 dimensional acoustic feature vectors are extracted as the basic data for subsequent process behavior recognition.

[0055] To achieve precise matching between acoustic signals and process behavior, processing instruction data needs to be extracted from the equipment control system or MES system, and the processing process needs to be divided and labeled into time periods. Processing instruction data includes the start time, processing duration, processing parameters (such as feed rate, load, and rotational speed), and product number for each process. This information can be extracted in real time via the OPC UA interface or PLC data reading module and aligned with the timestamp of the acoustic acquisition system, with a synchronization error controlled within ±0.1 seconds. By parsing the equipment processing logs, the system divides continuous acoustic monitoring data into several specific process segments, such as "rough grinding start-up period," "rough grinding stabilization period," and "polishing start-up period," forming a structured "crystal processing stage division map." The map records the process actions, equipment status, and operating parameters corresponding to each time period, guiding subsequent acoustic signature extraction and behavioral trend modeling. In practical applications, comparison of the time-series segment divisions of 50 batches of crystal processing data verified that the map accurately reflects the actual process flow, has good generalization and traceability capabilities, and provides a data foundation for achieving full-process monitoring and anomaly localization. After obtaining the complete processing stage map and time-frequency response characteristics of each stage, it is necessary to construct the acoustic signature behavior trajectory of the crystal in each processing stage to achieve dynamic modeling and anomaly identification of its acoustic state. This step first performs dimensionality reduction on the feature vectors extracted from each stage, using Principal Component Analysis (PCA) to compress the original 64-dimensional features into a 3-dimensional trajectory space, and then combines this with the t-SNE algorithm for clustering visualization, facilitating observation of the evolutionary trends between different behavior patterns. Subsequently, a Hidden Markov Model (HMM) is used to model the acoustic state of each stage in the processing, setting the number of states to 5, training the state transition probability matrix, and outputting the most probable state sequence to form the "crystal acoustic signature behavior trajectory." Abrupt points, stagnant states, or atypical migration paths in the trajectory are considered potential anomaly signals. In actual testing, complete trajectory data of 300 crystals were collected. Comparison with finished product test data revealed that crystals with obvious state abrupt changes in the acoustic signature trajectory generally had a final Q value more than 30% lower than the standard value. Therefore, this acoustic behavior trajectory analysis method can not only realize online status monitoring, but also provide early warning of equipment abnormalities or processing deviations, providing key technical support for production process optimization and product quality control.

[0056] In this embodiment, the specific steps for performing environmental noise adaptive filtering on the micro-acoustic response signal to generate a filtered and optimized response signal are as follows:

[0057] The micro-acoustic response signal is subjected to non-periodic mechanical sound identification to mark the mechanical noise in the production environment;

[0058] The mechanical noise in the production environment is located, and multiple noise source locations are marked.

[0059] Identify the spatial noise distribution of multiple noise source locations and construct a noise source distribution map of the production environment.

[0060] Based on the distribution map of noise sources in the production environment, the propagation time delay, phase drift, and energy gain curvature are analyzed to obtain multi-dimensional characteristics of the environmental noise signal;

[0061] Adaptive filtering optimization is performed based on the multi-dimensional characteristics of environmental noise signals to generate a filtered and optimized response signal.

[0062] In this embodiment, during the production process of quartz crystal resonators, non-periodic mechanical noise interference is prevalent in the workshop environment, such as the instantaneous impact sound of crane rails, the starting sound of auxiliary equipment, and the occasional sound of tools falling. These noise signals are typically characterized by strong suddenness, short duration, and wide spectral distribution, which can severely interfere with the identification of weak acoustic vibration features. To achieve the identification of non-periodic noise, it is first necessary to separate the acoustic components unrelated to equipment processing from the original micro-acoustic response signal. This step adopts a method based on a combination of statistical mutation detection and feature template matching for non-periodic mechanical noise identification. The specific process is as follows: short-time energy envelope extraction and spectral entropy analysis are performed on the original acoustic signal to identify signal segments with short-time energy mutations and sudden increases in spectral entropy values ​​as candidate non-periodic events; subsequently, the constructed mechanical noise feature library (containing 15 typical noise samples such as fan switching sound and pneumatic device impact sound) is introduced, and dynamic time warping (DTW) is used to perform time-frequency morphological matching between candidate events and template samples. In the experiment, a training sample set was constructed using 60 hours of production data, achieving an accuracy rate of over 92%. After processing, the system automatically identifies non-processing-related transient interference segments in the acoustic signal stream and removes these segments for subsequent filtering optimization, significantly improving the purity and reliability of the equipment's acoustic response. To further enhance the spatial resolution of noise recognition, it is necessary to accurately locate the specific physical locations of non-periodic noise within the workshop, thereby constructing a spatial labeling system for environmental noise. This step employs Time Difference of Arrival (TDOA) array-based microphone localization technology to locate the sound source, combined with a spatial inversion algorithm for multi-source location calibration. Specifically, nine high-sensitivity digital microphone array nodes (such as the GRAS 40PH type) are deployed within the workshop in a triangular and rhomboid arrangement, with node spacing controlled between 2-4 meters and a sampling rate set to 48kHz, all uniformly controlled by a synchronous acquisition main control system clock. When a non-periodic noise event occurs, each node acquires the received waveform data, and the arrival time difference between nodes is calculated through cross-correlation. The relative distance between the sound source and the array nodes was calculated using a least-squares inversion model, and then geometrically located using a workstation map. In the experiment, five known noise sources in the workshop (fans, elevators, gate controls, air compressors, and manual tapping points) were located, with the location error controlled within ±0.25 meters. Each location result was recorded as a "noise source location point" with a sound source type label, providing data support for constructing subsequent noise distribution maps and laying the foundation for spatial shielding and directional filtering of specific noise components.

[0063] Based on the identified locations of multiple noise sources, a complete spatial distribution map of workshop noise needs to be constructed to visualize the relative positions, sound intensity distribution, and main radiation directions of each sound source in space. This will provide a reference for subsequent noise propagation modeling and directional noise reduction. This step employs a method of sound intensity vector field reconstruction and equivalent noise level distribution calculation. Specifically, the entire workshop is divided into 0.5m × 0.5m three-dimensional acoustic grid units. Using the instantaneous sound pressure level and sound intensity direction measured by each microphone array, the equivalent sound pressure level (Leq) and direction cosine within each grid are calculated using the superposition method. To compensate for the sound reflection and interference effects caused by high reflectivity surfaces (such as walls and metal surfaces of equipment), a simplified ray tracing model is introduced to simulate primary reflection and scattering behavior, and weighted adjustments are made for locally excessively high reflectivity areas. In the experiment, 1000 grid cells were deployed in a 15m×10m×3m workshop. The reconstructed sound field map clearly showed a significant sound intensity concentration area (>85dBA) near the compressor area on the east side and the feed gate on the south side, while the noise in the center of the operating area was relatively weak (<65dBA). The final output "Production Environment Noise Source Distribution Map" visualized the sound power, influence range, and main radiation direction of each noise source, providing a data foundation for precise filtering and regional noise compensation. After establishing the noise source distribution map, it is necessary to further model the propagation characteristics of sound waves in the complex workshop environment and analyze its multi-dimensional characteristics to reveal the path dependence and dynamic change law of noise interference on signals. This step mainly analyzes three key acoustic propagation parameters: propagation time delay, phase drift, and energy gain curvature. First, for each noise source point, the time delay of its sound wave reaching each sensor node is calculated to form a delay matrix; this delay reflects the differences in spatial geometric path and medium properties, and is the basis for subsequent phase adjustment. Secondly, the phase information of the received signal at each node was analyzed, and the phase drift sequence was extracted. The phase shift characteristics in different frequency bands were evaluated using Fast Fourier Transform (FFT) and instantaneous frequency analysis. In the experiment, it was found that the signal on the corner reflection path had the largest phase drift (>120°) in the 8-12kHz frequency band. Thirdly, the gain change rate on the energy propagation path was evaluated, and the energy concentration and diffusion trend in space was estimated using the envelope curvature calculation method, reflecting the noise enhancement or attenuation characteristics. After synthesizing the three-dimensional sound propagation modeling data, a multi-dimensional propagation feature vector was formed for each noise source, including delay, phase, and energy gradient information. This provides targeted parameter inputs for subsequent filtering algorithms and lays the foundation for constructing a real-time sound field change model. Furthermore, an adaptive filtering algorithm based on spatial dynamic modeling was constructed to achieve accurate noise suppression and structural protection of the target acoustic signal. Unlike traditional amplitude or frequency-based filtering methods, a multi-dimensional adaptive filtering model integrating spatial delay, phase drift, and energy path was constructed. This model introduces a propagation feature weighting function in the filter design, allowing the filter weights to adaptively change in time and space.Specifically, an adaptive filter network (LMS + phase drift correction module) based on the minimum mean square error criterion was constructed. In each iteration, the learning rate and filter coefficients were dynamically adjusted according to the noise source location and propagation characteristics. In experimental signal processing tests at different process stages, the average signal-to-noise ratio improved by approximately 9.2 dB after processing using this method, with particularly significant suppression of transient noise during equipment startup and shutdown. Furthermore, the spectral integrity index of the filtered signal improved by more than 17%, significantly enhancing the ability to capture weak micro-vibration events (such as crystal peeling and scratching sounds). The final filtered and optimized response signal is not only clearer and more recognizable but also provides crucial high-quality data support for subsequent time-frequency behavior analysis and defect modeling.

[0064] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include:

[0065] A multi-angle structured light interferometer array was deployed to perform acoustic monitoring and simultaneous visual acquisition, resulting in a sequence of light field interferograms corresponding to the processing time sequence.

[0066] Based on the processing instructions of the equipment, the processing parameters are identified, and the tangential direction and the preset crystal cutting thickness variation law are calculated.

[0067] Based on the tangential direction and the preset crystal cutting thickness variation law, a theoretical processing model is fitted, and the interference fringes distribution is simulated to construct a theoretical optical field interference fringes distribution model.

[0068] Based on the theoretical optical field interference pattern distribution model, the optical field interference pattern sequence is image registered time-by-time and depth difference calculation is performed to obtain the theoretical crystal morphology response anomaly region.

[0069] The optical texture response deviation is calculated for the abnormal region of the theoretical crystal morphology response, and the optical texture response deviation segment is extracted.

[0070] In this embodiment, a multi-angle structured light interferometer array is deployed at key process nodes (such as crystal cutting, grinding, and fine-tuning) for acquiring visual interference images under synchronous acoustic signal acquisition. The structured light interferometer used is a dual-beam interferometric optical system, employing a 532nm green laser. A semi-transparent mirror is used to split the beam, generating a reference beam and a measurement beam. The two beams form interference fringes on the surface of the crystal under test and its reflectors, recording their phase difference and morphological changes. Each interferometer is equipped with a high-speed CCD camera with a resolution of 2048×2048 and a sampling frame rate of no less than 200fps, ensuring accurate capture of the optical interference pattern at each processing stage during dynamic processing. The array is arranged in a 120° interval, three-way distribution, covering the top, bottom, left, right, and oblique sides of the processed crystal, meeting the observation requirements for areas with arbitrary curvature changes. The system interfaces with the acoustic acquisition unit through a synchronous trigger control module, achieving high alignment between the image acquisition timestamp and the acoustic event, with a maximum error of no more than 2ms. The acquisition cycle is dynamically set according to the equipment's processing rhythm. For example, an image sequence is acquired once per second during the cutting stage, and increased to once every 0.2 seconds during the fine-tuning stage. The resulting light field interferogram sequence forms a dataset that matches the process timing, laying a crucial visual foundation for subsequent anomaly identification and image registration processing. Based on real-time command data provided by the equipment's processing control system, key parameters such as cutting speed (typ. 2-4 mm / s), feed rate (typ. 0.1-0.3 mm / min), spindle speed (typ. 2000-3000 rpm), and tool wear status are analyzed. Dynamic thickness variation modeling is then performed by combining the crystal material (e.g., AT-cut quartz) and the preset target thickness (typ. 100-160 μm). Tangential variation modeling is mainly based on the cutting path (straight / curved), wafer rotation angle, and feed direction, which are then coordinate-processed to express the thickness variation function as a two-dimensional function h(t, θ) of time and path. In experimental verification, thickness analysis was performed on 30 continuously processed crystals using a profilometer, revealing a strong correlation between the actual thickness variation and the equipment commands (R0). 2 A value greater than 0.95 indicates that the command data can be used as effective input for tangential modeling. Ultimately, the thickness variation model output from this step is not only used to predict the morphological variation trend of the crystal, but also provides a theoretical basis for the thickness and displacement evolution for subsequent theoretical interferogram simulations.

[0071] A theoretical machining geometry model was established to simulate the evolution and distribution of interference fringes that may occur during machining. The model was constructed using the finite element method (FEM) combined with geometric optics simulation technology. Thickness perturbations caused by crystal cutting were transformed into changes in optical path difference, thereby predicting the density, orientation, and rate of change of interference fringes. The established model considered various influencing factors, including tool eccentricity, crystal thermal expansion coefficient (typically 13.7 × 10⁻⁶ / K), and the vibration frequency of the cutting surface. Interference fringe simulation was performed using Zemax optical design software. By inputting the changes in cutting morphology as a displacement perturbation function, the relationship between the interference optical path difference and phase superposition was calculated, and the fringe morphology evolution diagram was output. The model resolution was set to 0.5 μm / pixel, covering the entire wafer surface (typ. 25-50 mm diameter). Simulation results showed that different thickness gradient regions could lead to local abrupt changes or distortions in fringe density. The final constructed "theoretical optical field interference fringe distribution model" provides interference patterns under ideal conditions at different machining stages, providing a comparison benchmark for subsequent image registration and anomaly detection. This model can be dynamically updated and is suitable for predictive analysis under different process parameters, possessing good generalizability and field adaptability. The acquired optical field interferogram sequence is registered with the theoretical fringing model on a time-by-time basis, and potential anomalous regions are identified through differential calculation methods. Image registration employs a multi-scale phase correlation algorithm (Phase Correlation with Log-Polar Transform), achieving a registration accuracy better than 0.2 pixels, effectively handling minute shifts caused by vibration or viewing angle changes. After registration, pixel-level phase difference calculations are performed on each actual interferogram and its corresponding theoretical image to generate a "phase residual map." In the experiment, if the phase deviation of a continuous region in the residual map exceeds π / 3 or the fringing density change rate is greater than ±20%, it is marked as a "theoretical crystal morphology response anomalous region." Furthermore, to suppress spurious differences caused by illumination variations, local histogram normalization is introduced to ensure the stability and repeatability of the differential data. In a typical crystal processing experiment, by comparing the interferograms of 80 products, it was found that 12 of them showed increased fringe bending in the later stage of fine-tuning, and obvious striped abnormal areas were formed in the residual diagram. Subsequent verification showed that their frequency consistency shift reached ±150ppm.

[0072] Further quantitative analysis of the interference morphology in these regions was conducted to extract representative deviation segments, which were used to establish the correspondence between crystal morphology anomalies and processing defects. The calculation of morphology response deviation includes two core dimensions: angular deviation and gradient jump ratio. Specifically, firstly, isophase lines of the striations are extracted from the anomalous region, and the direction of the main striations is extracted using Hough transform. The angle is compared with the theoretical fringe direction; a deviation exceeding 5° is considered an "angular deviation segment." Secondly, the rate of change in the number of fringes per unit area is calculated; if the density change exceeds 30%, it is determined to be a "density deviation segment." These deviation segments are mapped by coordinates and timestamped to form a dataset of "optical morphology response deviation segments," which is used for comparative analysis with acoustic anomaly segments. Experimental samples showed that crystals exhibiting optical deviation segments had uneven surface morphology after subsequent electrode deposition, with a Q-value loss as high as 18%, proving that this optical anomaly identification method can effectively predict structural processing defects. This step enables a deep integration of the visual and acoustic domains, providing precise and multidimensional data support for intelligent monitoring and defect diagnosis in the quartz crystal resonator manufacturing process.

[0073] In this embodiment, reference Figure 4 The diagram below illustrates the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include:

[0074] Time-series synchronization and spatial mapping of multi-stage crystal acoustic waveform behavior trajectory sequences and optical waveform morphology response deviation segments were performed to establish a joint acoustic-optical spectrum of crystal production.

[0075] Acoustic-optic behavior frequency response analysis was performed on the acousto-optic combined spectrum of crystal production to obtain the time-shift characteristics of the acousto-optic behavior;

[0076] Based on the acousto-optic combined spectral analysis of crystal production, phase shift and spectral morphology are analyzed to obtain the spectral center drift and image interference peak misalignment;

[0077] A database of acoustic-optic fusion feature deviations was constructed by fitting the acoustic-optic mismatch index to the time shift characteristics of acoustic-optic behavior, spectral center drift, and image interference peak misalignment.

[0078] In this embodiment, the "multi-stage crystal acoustic signature behavior trajectory sequence" in the acoustic domain and the "optical morphology response deviation segment" in the visual domain are fused across modes. Through precise temporal synchronization and spatial coordinate mapping, a unified "crystal production acoustic-optical joint map" is constructed. First, based on a unified timestamp mechanism, high-precision clock synchronization modules (using the IEEE 1588 protocol with a synchronization accuracy of ±1ms) are embedded in both the acoustic monitoring system and the structured light interferometry vision system to ensure strict temporal alignment between acoustic signature data and optical field image acquisition. Then, a process timeline mapping relationship is established for the two types of data, and the phase divisions of the acoustic signature trajectory (such as rough grinding, polishing, and fine-tuning) are compared with the optical morphology anomaly segments (such as interference fringe distortion and abrupt morphological changes) through temporal overlap. In the spatial dimension, calibration is performed using the geometric coordinate system of the process tooling and fixtures. The two-dimensional coordinate system (unit: μm) of the interferometric image is mapped to the mechanical station coordinates of the acoustic signature monitoring point, and a rigid transformation matrix (including rotation and displacement parameters) is used to complete spatial reprojection matching. In practical applications, acousto-optic data fusion was performed on 30 crystal samples, revealing an average overlap rate exceeding 86%, with calibration deviation controlled within ±0.25 min. The resulting acousto-optic joint spectrum is a multi-channel dataset with time as the primary axis and spatial coordinates as an auxiliary dimension, supporting subsequent multi-level intelligent analysis tasks such as behavioral coupling analysis, time-frequency anomaly localization, and defect pattern recognition. Joint time-frequency analysis and cross-correlation analysis methods were employed to extract the frequency response trends of acoustic features and optical ripple responses, and to calculate the time offset characteristics between acoustic and optical behaviors. First, the acoustic trajectory was segmented into multiple stages, and the dominant frequency components (typically 3-15 kHz) of each stage were extracted. Then, the pixel intensity time-domain transformation was performed on the image change sequence corresponding to the optical ripple anomaly segment, and the intensity variation of the ripple morphology at different frequencies was extracted using two-dimensional image Fourier transform (2D-FFT). Finally, the cross-correlation function between the acoustic energy envelope and the image intensity change curve was calculated to obtain the time delay τ between them, which was defined as the "acoustic-optic behavior time offset characteristic." In the experiment, analysis of 120 sets of combined acousto-optic spectra revealed that the acousto-optic response delay was controlled within ±80ms in most stable processing stages. However, in the initial stage of crystal cutting and the final stage of fine-tuning, the delay could exceed 200ms, reflecting the non-uniform acousto-optic behavior caused by process disturbances. This type of time shift characteristic is of great value for early anomaly prediction, especially in scenarios such as thermal cutting, tool noise, or a surge in vibration, where changes in optical morphology often lag behind the acoustic response, making acousto-optic time shift one of the key indicators of early warning signals.

[0079] Phase and spectral morphology analysis was performed on the data from each stage of the acoustic-optical joint spectrum to extract two key feature indicators: "spectral center drift" and "image interference peak misalignment." In the acoustic signal section, stable periodic segments in the acoustic traces of each processing stage were selected and analyzed using Fast Fourier Transform (FFT) to extract their spectral center frequency, which was then differentially calculated with the previous time period. If the center frequency drift exceeded ±400Hz, it was marked as a "spectral center drift segment." Secondly, in the optical spectrum section, phase unfolding was performed on each frame of the interference image to extract the position of the interference peak (i.e., the maximum phase bright band), and error matching was performed to align it with the theoretical model. If the peak position shift exceeded 10μm, it was defined as "image interference peak misalignment." In typical processing, it was found that when the tool wear approached a critical state, the acoustic center drifted from 7.2kHz to 8.1kHz, and simultaneously, nonlinear bending and misalignment of the interference peaks in the image occurred. These two phenomena showed a high degree of overlap in the timeline (cross-correlation coefficient reached 0.87), indicating that this type of feature has strong coupling and high diagnostic value. The two types of feature indicators extracted in this step not only help reveal the physical behavior behind the acoustic-optical response but also provide core input dimensions for constructing the fusion model. Using the acoustic-optical behavior time offset (Δt), acoustic spectrum center shift (Δf), and image interference peak shift (Δp) as three-dimensional input variables, an acoustic-optical mismatch index vector [Δt, Δf, Δp] is constructed. Then, historical sampling data is classified and aggregated based on the K-means++ clustering algorithm, defining multiple deviation types, such as "mild synchronization mismatch," "frequency exacerbation," "interference peak misalignment," and "strong coupling mismatch." Each cluster center represents a typical acoustic-optical deviation pattern, and a label index is established in conjunction with equipment parameters and process stages. In the experiment, nearly 800 sets of acoustic-optical joint data samples were collected and archived. After deviation fitting and pattern classification, nine high-frequency acoustic-optical coupling deviation patterns were summarized, each associated with specific equipment status, process fluctuations, or material anomalies. This database can be connected to MES systems and equipment predictive maintenance platforms to achieve real-time comparison and deviation alarms, and can also be used for offline training of AI defect diagnosis models. Ultimately, this acoustic-optical fusion deviation database not only serves as a knowledge hub for production monitoring, but also provides important data support and decision-making basis for cross-process collaborative optimization.

[0080] In this embodiment, step S4 includes the following steps:

[0081] A three-dimensional spatial mapping was performed on the acoustic-optic fusion feature deviation database to construct a three-dimensional acoustic-optic mapping model.

[0082] Based on the three-dimensional acousto-optic mapping model, the time delay difference of sound wave propagation is calculated in reverse, and the time delay difference inverse calculation information is extracted.

[0083] Light field imaging mining is performed on the three-dimensional acousto-optic mapping model, and the location of three-dimensional defect points is determined by back-calculation of time delay difference information, resulting in accurate location coordinates of multiple three-dimensional defect points;

[0084] The acoustic reflection morphology and optical interference morphology of the three-dimensional defect points are analyzed together to obtain the defect point type characteristics.

[0085] In this embodiment, the high-dimensional feature vectors [Δt, Δf, Δp] extracted from the acousto-optic fusion feature deviation database (representing the time offset of acousto-optic behavior, the drift of the acoustic spectrum center, and the displacement of the image interference peak, respectively) are projected onto the physical processing space to construct a three-dimensional acousto-optic mapping model that can be used for localization and defect behavior interpretation. First, based on the mechanical coordinate system (XYZ three-axis system, unit accuracy of 0.01mm) of the crystal processing platform and the completed acousto-optic joint spectrum spatial mapping relationship, each set of acousto-optic mismatch vectors is spatially aligned with the specific spatial position of the crystal during processing (such as cutting path points, polishing trajectory lines, or fixed fixture coordinates). This process is completed collaboratively using the spindle processing trajectory tracking method and the visual positioning point calibration method, and all acousto-optic feature data are labeled with their positions. To ensure mapping accuracy, in experimental verification, the interference spectra and acoustic responses of 50 crystals at fixed reference positions are compared, and the mapping deviation is controlled within 0.12mm. Subsequently, a point cloud voxel reconstruction method (Voxel-based 3D mapping) was used to map all processing events with acousto-optic deviation characteristics into a three-dimensional feature point array, forming a three-dimensional map of "acousto-optic behavior-spatial configuration". This three-dimensional acousto-optic mapping model not only preserves the acoustic-optical behavior intensity of each processing node, but also integrates multi-dimensional labels such as process cycle information and equipment load status, which can directly support subsequent defect point tracing and predictive diagnosis tasks based on spatial-behavioral coupling. Focusing on the propagation path and delay difference analysis of sound waves inside the crystal, the "time delay difference back-calculation information" of sound wave propagation was extracted. This information serves as "inversion data" for the acoustic anomaly propagation mechanism inside the crystal and can be used to assist in spatial positioning and defect physical property identification. Specifically, multiple acousto-optic coupling anomaly points are selected in the acousto-optic mapping model, their acoustic response waveform data are extracted, and the arrival time difference Δt′ of the main wave packet between the point and the reference point (usually the processing standard area under no-anomaly state) is calculated. Then, a theoretical propagation time model was established based on the intrinsic sound velocity of quartz (approximately 5760 m / s for longitudinal waves and approximately 3330 m / s for transverse waves in AT-cut quartz), and the propagation path was inverted using a crystal spatial geometry model. Beam tracing was employed to simulate the propagation path of sound waves in a non-uniform crystal, and the difference between the theoretical propagation time and the actual observed time was compared to calculate a "time delay offset map." In the experiment, a laser ultrasonic excitation device was used to excite the crystal at different locations at a frequency of 10 MHz. Combined with signal acquisition from six array sensors, the final inverted time delay difference ranged from ±4 μs to ±45 μs. This time delay difference reflects the local variations in parameters such as the density and elastic modulus of the internal medium of the crystal, and is an important precursor indicator of the presence of microcracks, inclusions, or stress concentration zones in the crystal, providing a fundamental calculation basis for the next step of precise three-dimensional defect location.

[0086] High-density reconstruction of the labeled anomalous spectral regions was performed using light field imaging technology. A multi-view structured light interferometer was used to acquire multi-angle interference image sequences of the crystal surface, and a voxel-based light field reconstruction algorithm was employed to construct a complete three-dimensional interference topography (spatial resolution reaching 0.5 μm / voxel). Next, spatial fusion analysis was performed between the acoustic delay difference-derived path and the three-dimensional light field map, comparing the paths of all locations exhibiting significant acoustic delay anomalies (Δt′>30 μs). By focusing the analysis of the intersection areas, three-dimensional spatial points meeting the following conditions were identified as "suspected defect points": 1) intersection with the anomalous acoustic propagation path; 2) overlap with the anomalous fringe aggregation area in the interferogram exceeding 85%; 3) high deviation amplitude (Δp>15 μm) and high spectral perturbation (Δf>500 Hz). Using this triple-criteria matching method, 0-3 "defect point coordinates" on each crystal are finally determined, with each point stored in the form of [X, Y, Z], with an accuracy better than ±0.08 mm. In actual crystal defect verification experiments, this method successfully located microcrack regions and buried inclusion regions, with a verification accuracy exceeding 92%, providing a core physical basis for subsequent classification, identification, and response strategy formulation. Further joint analysis of the acoustic reflection behavior and optical interferogram morphology of each defect point is conducted to extract its type characteristics, used to distinguish the physical nature of the defects (such as cracks, inclusions, heterogeneous layers, residual stress concentration, etc.). First, in the acoustic dimension, a short-time Fourier transform (STFT) is performed on the reflection waveform of each defect point in the original acoustic response data to extract the amplitude and phase spectrum characteristics of the reflected wave. Simultaneously, attention is paid to the presence of typical morphologies such as "scattering tail waves" and "multiple reflection peaks" in the reflected wave packets as indicators of structural discontinuities or the presence of interfaces. In the experiment, typical crack defects exhibited a delayed scattering peak of 0.2-0.4 ms, while inclusion defects showed abrupt changes in high-frequency reflection. Secondly, in the optical dimension, the fringe morphology shift patterns in the corresponding interference image regions were analyzed, with particular attention paid to changes in fringe torsion rate, density gradient, and symmetry. Crack defects often manifested as local fringe breakage and nonlinear shift, while inclusions caused outward radial expansion or elliptical ring-shaped extension of the fringes. Finally, the acoustic reflection morphology and optical fringe deviation morphology were fused, and a "defect type discrimination vector" was constructed using fuzzy rules or a neural network (which can be implemented in subsequent modules), outputting defect type labels (such as Crack, Inclusion, Delamination, Stress Concentration, etc.).The joint identification method achieved a classification accuracy of 91.3% in laboratory comparative sample tests, which is significantly better than single acoustic or optical methods. This marks a new stage of higher accuracy and interpretability for defect identification based on three-dimensional fusion models, providing a practical and intelligent support path for real-time defect diagnosis of quartz crystal resonators.

[0087] In this embodiment, the specific steps of step S5 are as follows:

[0088] A spatial label set is established based on the precise location coordinates of multiple 3D defect points, and the defect location coordinates, type attributes and formation time index parameters of each stage are extracted.

[0089] Based on the defect point type characteristics, defect location coordinates at each stage, type attributes and formation time index parameters, a time series analysis of time series deviation is performed to construct a cumulative deviation time axis.

[0090] Dynamic evolution modeling is performed based on the deviation accumulation time axis, and the final deviation accumulation prediction is performed to generate the final deviation accumulation result.

[0091] The final accumulated deviation is analyzed for time-series deviation evolution to generate structural deviation trend curves and expected electrical performance index drift values.

[0092] In this embodiment, for the defect point coordinate data obtained from the 3D acousto-optic mapping model, each defect point is accurately located and recorded using the spatial coordinate system (XYZ axes, resolution 0.01mm) of the crystal processing platform. In addition to spatial coordinates, each defect point is also associated with defect type attributes obtained through acousto-optic joint identification, such as category information like crack, inclusion, and stress concentration. Simultaneously, combined with the real-time processing data acquisition system, the time index parameter of defect formation is extracted, i.e., the time point when the defect first appears or anomaly is detected (accurate to the second). This time index relies on the equipment processing command sequence and sensor sampling timestamps for synchronous annotation, ensuring high-precision correspondence between spatiotemporal data. Through the above processing, a multi-dimensional spatial label set is established, including 3D position (X, Y, Z), defect type (e.g., Crack, Inclusion), formation time index (t0), and processing stage information (e.g., rough grinding, polishing). To ensure the integrity and accuracy of the label set, a data verification mechanism is adopted, such as verifying that the coordinate deviation is less than 0.05mm through repeated scanning, achieving a type determination consistency rate of 92%. This spatial label set provides high-quality basic data for subsequent time-series deviation analysis and dynamic modeling. Defect point data were sorted by time series according to the defect formation time index parameter to form a time-series event stream. For each defect type, the time points and frequencies of its occurrence in each process stage (such as cutting, polishing, etc.) were statistically analyzed, and a curve of defect generation frequency versus time was plotted. The sliding window method was used to aggregate the number of defects and abnormal indicators over time to analyze the defect accumulation trend and fluctuation pattern. Specifically, by calculating the hourly increase in the number of defects and the corresponding rate of change of acoustic and optical characteristics, a cumulative deviation function P(t) was constructed to reflect the defect development speed and the stage acceleration phenomenon. In the experiment, for data from a typical crystal production line, it was found that crack-type defects began to accumulate from the initial polishing stage, with an average cumulative rate increase of 0.15 points per hour, while inclusion-type defects appeared frequently in the rough grinding stage, with a cumulative rate reaching 0.25 points. In addition, by comparing the spatial coordinates of defects, spatial clustering of some defects was observed, and the spatiotemporal clustering algorithm (ST-DBSCAN) was further used to identify local deviation hotspots. Finally, the deviation accumulation time axis fully shows the dynamic process of various defects from their appearance to their development, providing an accurate temporal basis for the deviation evolution model.

[0093] Nonlinear time series modeling methods, such as a hybrid algorithm based on Long Short-Term Memory (LSTM) networks and Autoregressive Moving Average (ARIMA) models, were employed to fit the number of defects to anomaly indices in acoustic and optical behavior, using accumulated deviation timeline data. Model inputs included historical accumulated defect values, defect type weights (assigned according to the level of damage), and changes in processing parameters (such as cutting speed and temperature fluctuations). During training, 700 processing cycle data points from the past three months were used, divided into training and validation sets, with the average prediction error on the validation set controlled within ±3%. The model dynamically captures the nonlinear acceleration effect and phased mitigation trends during the deviation accumulation process. By inferring future time points, the model outputs the final accumulated deviation result, represented by a predicted map of the total defect growth and spatial distribution over the next few hours to days. Furthermore, Monte Carlo simulation was used to calculate the uncertainty interval of the model's predictions, enhancing the robustness of the predictions. The final results are presented in a digital report format, including the accumulated deviation curve, a predicted spatial heatmap, and a risk level assessment, providing a scientific basis for production scheduling and maintenance decisions. By utilizing historical production data collected from the factory, a regression model was established between structural deviation indices (such as crack length and inclusion density) and electrical performance parameters (such as resonant frequency drift and Q-factor reduction). Combined with the cumulative deviation curve, the evolution trend of structural deviations over a future period was predicted. Secondly, multivariate regression and principal component analysis (PCA) were used to summarize various defect types and their corresponding electrical performance drifts into a few dominant factors. Experimental verification showed that when the cumulative deviation index exceeded 0.6 (normalized value), the resonant frequency drift exceeded 5 ppm, and the Q-factor decreased by more than 10%. Subsequently, based on the predicted cumulative deviation data, a time-series structural deviation trend curve was generated, reflecting the impact of dynamic defect growth on product electrical performance. This curve was presented through a visual interface, assisting engineers in real-time assessment of production line status and adjustment of processing parameters. Finally, combined with preset electrical performance thresholds, the system automatically triggered early warnings, indicating potential functional failure risks and ensuring stable production quality. This trend evolution analysis not only enables real-time defect monitoring but also provides targeted electrical performance risk prediction capabilities, significantly improving the intelligence level and product reliability of quartz crystal resonators.

[0094] In this embodiment, step S6 is as follows:

[0095] Based on the structural deviation trend curve and the expected electrical performance index drift value, a multidimensional comparison of the theoretical finished product is carried out to calculate the morphological residual distribution and frequency drift range.

[0096] Deviation threshold assessment is performed based on morphological residual distribution and frequency drift interval, and a comprehensive deviation assessment is conducted to generate a deviation assessment report.

[0097] Based on the deviation assessment report, a performance impact diagnosis is performed to obtain the estimated performance impact results under production defects;

[0098] Based on the performance prediction impact results, iterative closed-loop control optimization of the production line is performed to carry out production defect diagnosis operations.

[0099] In this embodiment, the aforementioned structural deviation trend curve and electrical performance index drift values ​​are cross-compared in multiple dimensions to deeply quantify the correspondence between product morphology changes and resonant frequency drift. Registration and comparison with the theoretical design model yields a morphology residual distribution map. Morphology residuals typically reflect local surface morphology deviations with micron-level precision, such as depressions, protrusions, and crack depths. Subsequently, these residual data are matched with corresponding frequency drift intervals, which are acquired using high-precision frequency measurement equipment, typically with a resolution of 1 ppm, accurately quantifying the range of crystal resonant frequency variation. Statistical methods, such as correlation coefficient calculation and regression analysis, are used to analyze the spatial correspondence and numerical correlation between morphology residuals and frequency drift. Experiments show that when the morphology residual exceeds 0.5 μm, the frequency drift often enters a fluctuation range of 3-7 ppm, and some areas have concentrated morphology defects, resulting in more severe frequency drift fluctuations. Furthermore, combining time-series trend data further confirms the spatiotemporal synchronicity between morphology anomaly areas and frequency drift peaks, providing a basis for subsequent deviation threshold setting. Based on historical process data and quality standards, upper and lower thresholds for morphological residuals and frequency drift are defined. For example, the morphological residual threshold is set to 0.4 μm, and the frequency drift threshold is 5 ppm. In actual production monitoring, a threshold judgment algorithm is used to identify out-of-limit areas and abnormal fluctuation periods based on the collected residual distribution map and frequency drift data. Further, a weighted scoring method is used to weight different deviation indicators, and a comprehensive deviation score is calculated. During this process, a fuzzy comprehensive evaluation method is used to integrate multiple parameters, enhancing the objectivity and comprehensiveness of the deviation assessment. A deviation score exceeding a preset threshold is considered a potential quality risk. The assessment results are summarized in charts and text, including a deviation hotspot area location map, an out-of-limit event statistics table, and a trend analysis curve. In experimental verification, this assessment process was used to analyze nearly 100 batches of crystal products, accurately capturing more than 85% of quality anomalies and significantly improving defect identification efficiency. Finally, the deviation assessment report generated by the system provides detailed data support for subsequent performance impact diagnosis.

[0100] Based on the previously established deviation-performance mapping model, various deviation indicators (mean and maximum values ​​of morphological residuals, peak frequency drift, etc.) from the deviation assessment report are input. Using multiple linear regression and support vector machine (SVM) algorithms, the corresponding changes in electrical performance indicators are predicted, such as resonant frequency shift, quality factor decrease ratio, and frequency stability indicators. Performance impact diagnosis also includes sensitivity analysis of different defect types to determine which deviations have the most critical impact on performance. Furthermore, by combining process stage information, the differences in the impact of defect formation timing on performance are analyzed. For example, cracks formed in the early polishing stage have a more significant impact on the Q value, while surface roughness fluctuations in the final cutting stage have a greater impact on frequency stability. In the experiment, by diagnosing 500 sets of product data, the diagnostic results showed a 92% agreement with the actual factory performance test results, demonstrating high accuracy and reliability. This performance prediction result can not only provide early warning of potential performance degradation but also guide production adjustments and maintenance priority ranking, improving overall production quality. The performance prediction results are input into the production management system (such as MES) and equipment control unit, automatically triggering adjustment actions through process control rules, including cutting speed adjustment, polishing pressure fine-tuning, and temperature and humidity control optimization. The system employs a real-time monitoring feedback loop, combined with sensor data from key equipment, to dynamically adjust processing parameters and reduce the probability of defects. Secondly, a defect diagnosis workflow is established, clearly defining the abnormal alarm response mechanism and manual intervention process to ensure rapid location and correction of process deviations in case of anomalies. To verify the effectiveness of closed-loop control, multiple batches of trial production experiments were conducted, comparing defect rates and performance index fluctuations before and after iterations. Experimental data shows that after closed-loop optimization, the defect rate decreased by approximately 20%, and frequency drift fluctuations decreased by 15%. Furthermore, the system supports data archiving and historical analysis, continuously optimizing the control model and improving the intelligence level of the control strategy. This closed-loop control not only achieves real-time defect diagnosis in the production process but also effectively ensures stable product performance and improved production efficiency, providing a solid guarantee for the intelligent and high-quality manufacturing of quartz crystal resonators.

[0101] In this embodiment, a manufacturing defect diagnosis system for quartz crystal resonators is provided, used to perform the manufacturing defect diagnosis method for quartz crystal resonators as described above, including:

[0102] The acoustic analysis module is used to extract the micro-acoustic response signals of each processing stage based on the ultrasonic excitation and response monitoring device, perform acoustic time-frequency structure analysis and response trajectory trend analysis and extraction, so as to obtain a multi-stage crystal acoustic pattern behavior trajectory sequence.

[0103] The optical morphology response module is used to deploy a multi-angle structured light interferometer array for acoustic monitoring and synchronous visual acquisition, calculate the optical morphology response deviation, and extract the optical morphology response deviation segment.

[0104] The acoustic-optic mismatch module is used to perform acoustic-optic behavior frequency response analysis on the multi-stage crystal acoustic pattern behavior trajectory sequence and optical pattern morphology response deviation segment, and to fit the acoustic-optic mismatch index to construct an acoustic-optic fusion feature deviation database.

[0105] The defect point localization module is used to locate three-dimensional defect points based on the acoustic-optical fusion feature deviation database, and to perform joint analysis of acoustic reflection morphology and optical interference morphology to obtain defect point type characteristics.

[0106] The deviation situation evolution module is used to perform time-series deviation analysis based on the characteristics of defect point type, and to perform time-series deviation situation evolution analysis to generate structural deviation trend curves and expected electrical performance index drift values.

[0107] The performance diagnostic module is used to diagnose the performance impact based on the structural deviation trend curve and the expected electrical performance index drift value, and to perform iterative closed-loop control optimization of the production line to execute production defect diagnosis operations.

[0108] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0109] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for diagnosing manufacturing defects in quartz crystal resonators, characterized in that, Includes the following steps: Step S1: Extract the micro-acoustic response signals of each processing stage based on the ultrasonic excitation and response monitoring device, perform acoustic time-frequency structure analysis and response trajectory trend analysis and extraction to obtain a multi-stage crystal acoustic behavior trajectory sequence; specifically: identify the quartz crystal resonator production line processing flow and mark key processing equipment; The key processing equipment is monitored for acoustic vibration in real time using an ultrasonic excitation and response monitoring device, and the micro-acoustic response signals of each processing stage are extracted. The micro-acoustic response signal is subjected to environmental noise adaptive filtering to generate a filtered optimized response signal; specifically, the micro-acoustic response signal is subjected to non-periodic mechanical sound identification to mark the mechanical noise of the production environment; The mechanical noise in the production environment is located, and multiple noise source locations are marked. Identify the spatial noise distribution of multiple noise source locations and construct a noise source distribution map of the production environment. Based on the distribution map of noise sources in the production environment, the propagation time delay, phase drift, and energy gain curvature are analyzed to obtain multi-dimensional characteristics of the environmental noise signal; Adaptive filtering optimization is performed based on the multi-dimensional characteristics of environmental noise signals to generate a filtered optimization response signal; Acoustic time-frequency structure analysis is performed on the filtered optimized response signal to generate time-frequency response characteristics for multiple processing stages; Obtain equipment processing instructions, divide time segments based on the equipment processing instructions, and generate a crystal processing stage division map; Based on the crystal processing stage division map, the acoustic behavior response trajectory trend analysis of the time-frequency response characteristics is performed and extracted to obtain a multi-stage crystal acoustic waveform trajectory sequence. Step S2: Deploy a multi-angle structured light interferometer array to perform acoustic monitoring and synchronous visual acquisition, calculate the light ripple morphology response deviation, and extract the light ripple morphology response deviation segment; Step S3: Perform acoustic-optical behavior frequency response analysis on the multi-stage crystal acoustic pattern behavior trajectory sequence and optical pattern morphology response deviation segment, and fit the acoustic-optical mismatch index to construct an acoustic-optical fusion feature deviation database; specifically: perform temporal synchronization and spatial mapping on the multi-stage crystal acoustic pattern behavior trajectory sequence and optical pattern morphology response deviation segment to establish a crystal production acoustic-optical joint spectrum. Acoustic-optic behavior frequency response analysis was performed on the acousto-optic combined spectrum of crystal production to obtain the time-shift characteristics of the acousto-optic behavior; Based on the acousto-optic combined spectral analysis of crystal production, phase shift and spectral morphology are analyzed to obtain the spectral center drift and image interference peak misalignment; A database of acoustic-optic fusion feature deviations was constructed by fitting acoustic-optic mismatch indices to the time shift characteristics of acoustic-optic behavior, spectral center drift, and image interference peak misalignment. Step S4: Based on the acoustic-optical fusion feature deviation database, locate the three-dimensional defect points and perform joint analysis of acoustic reflection morphology and optical interference morphology to obtain the defect point type characteristics; Step S5: Perform time series analysis on the defect point type characteristics and analyze the evolution of the time series deviation to generate the structural deviation trend curve and the expected electrical performance index drift value. Step S6: Based on the structural deviation trend curve and the expected electrical performance index drift value, perform performance impact diagnosis and iterative closed-loop control optimization of the production line to execute production defect diagnosis.

2. The method for diagnosing manufacturing defects in quartz crystal resonators according to claim 1, characterized in that, The specific steps of step S2 are as follows: A multi-angle structured light interferometer array was deployed to perform acoustic monitoring and simultaneous visual acquisition, resulting in a sequence of light field interferograms corresponding to the processing time sequence. Based on the processing instructions of the equipment, the processing parameters are identified, and the tangential direction and the preset crystal cutting thickness variation law are calculated. Based on the tangential direction and the preset crystal cutting thickness variation law, a theoretical processing model is fitted, and the interference fringes distribution is simulated to construct a theoretical optical field interference fringes distribution model. Based on the theoretical optical field interference pattern distribution model, the optical field interference pattern sequence is image registered time-by-time and depth difference calculation is performed to obtain the theoretical crystal morphology response anomaly region. The optical texture response deviation is calculated for the abnormal region of the theoretical crystal morphology response, and the optical texture response deviation segment is extracted.

3. The method for diagnosing manufacturing defects in quartz crystal resonators according to claim 1, characterized in that, The specific steps of step S4 are as follows: A three-dimensional spatial mapping was performed on the acoustic-optic fusion feature deviation database to construct a three-dimensional acoustic-optic mapping model. Based on the three-dimensional acousto-optic mapping model, the time delay difference of sound wave propagation is calculated in reverse, and the time delay difference inverse calculation information is extracted. Light field imaging mining is performed on the three-dimensional acousto-optic mapping model, and the location of three-dimensional defect points is determined by back-calculation of time delay difference information, resulting in accurate location coordinates of multiple three-dimensional defect points; The acoustic reflection morphology and optical interference morphology of the three-dimensional defect points are analyzed together to obtain the defect point type characteristics.

4. The method for diagnosing manufacturing defects in quartz crystal resonators according to claim 1, characterized in that, The specific steps of step S5 are as follows: A spatial label set is established based on the precise location coordinates of multiple 3D defect points, and the defect location coordinates, type attributes and formation time index parameters of each stage are extracted. Based on the defect point type characteristics, defect location coordinates at each stage, type attributes and formation time index parameters, a time series analysis of time series deviation is performed to construct a cumulative deviation time axis. Dynamic evolution modeling is performed based on the deviation accumulation time axis, and the final deviation accumulation prediction is performed to generate the final deviation accumulation result. The final accumulated deviation is analyzed for time-series deviation evolution to generate structural deviation trend curves and expected electrical performance index drift values.

5. The method for diagnosing manufacturing defects in quartz crystal resonators according to claim 1, characterized in that, The specific steps of step S6 are as follows: Based on the structural deviation trend curve and the expected electrical performance index drift value, a multidimensional comparison of the theoretical finished product is carried out to calculate the morphological residual distribution and frequency drift range. Deviation threshold assessment is performed based on morphological residual distribution and frequency drift interval, and a comprehensive deviation assessment is conducted to generate a deviation assessment report. Based on the deviation assessment report, a performance impact diagnosis is performed to obtain the estimated performance impact results under production defects; Based on the performance prediction impact results, iterative closed-loop control optimization of the production line is performed to carry out production defect diagnosis operations.

6. A manufacturing defect diagnosis system for quartz crystal resonators, characterized in that, A method for performing manufacturing defect diagnosis of a quartz crystal resonator as described in claim 1, comprising: The acoustic analysis module is used to extract the micro-acoustic response signals of each processing stage based on the ultrasonic excitation and response monitoring device, perform acoustic time-frequency structure analysis and response trajectory trend analysis and extraction, so as to obtain a multi-stage crystal acoustic pattern behavior trajectory sequence. The optical morphology response module is used to deploy a multi-angle structured light interferometer array for acoustic monitoring and synchronous visual acquisition, calculate the optical morphology response deviation, and extract the optical morphology response deviation segment. The acoustic-optic mismatch module is used to perform acoustic-optic behavior frequency response analysis on the multi-stage crystal acoustic pattern behavior trajectory sequence and optical pattern morphology response deviation segment, and to fit the acoustic-optic mismatch index to construct an acoustic-optic fusion feature deviation database. The defect point localization module is used to locate three-dimensional defect points based on the acoustic-optical fusion feature deviation database, and to perform joint analysis of acoustic reflection morphology and optical interference morphology to obtain defect point type characteristics. The deviation situation evolution module is used to perform time-series deviation analysis based on the characteristics of defect point type, and to perform time-series deviation situation evolution analysis to generate structural deviation trend curves and expected electrical performance index drift values. The performance diagnostic module is used to diagnose the performance impact based on the structural deviation trend curve and the expected electrical performance index drift value, and to perform iterative closed-loop control optimization of the production line to execute production defect diagnosis operations.

Images