A monitoring method and system based on quartz crystal cutting anomaly

Abstract

The present invention discloses a monitoring method and system based on quartz crystal cutting anomalies, which relates to the field of anomaly monitoring technology, including: using a high-speed industrial camera and a laser scanning spectrometer to capture the surface image and reflection spectrum of each section during the cutting process; pre-processing the surface image and the reflection spectrum and then performing spatial registration; performing feature fusion of the spatially registered image and the surface cutting path to obtain a clear input data stream; performing pyramid reconstruction on the data stream corresponding to each section; simplifying the inherent parameters of the reconstructed section data stream, analyzing the simplified section data stream, and obtaining the detection results of quartz crystal section anomalies. This method can significantly improve the automation and intelligence level of anomaly detection in the quartz crystal cutting process, breaking through the technical bottleneck of traditional methods' weak ability to identify small defects and low-contrast anomalies.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormality monitoring, and in particular to a monitoring method and system based on quartz crystal cutting abnormalities. Background Art

[0002] Due to their excellent physical properties and chemical stability, quartz crystals are widely used in high-end manufacturing applications such as electronic components, precision instruments, and optical materials. In these applications, quartz crystals typically undergo rigorous cutting, polishing, and forming processes to achieve device structures that meet performance requirements. However, the quartz crystal cutting process is extremely complex, limited not only by the anisotropy and brittleness of the material itself, but also by multiple factors such as the processing environment, tool condition, and process parameters. Because quartz materials are prone to defects such as microcracks, chipping, and surface scratches, any minor processing anomaly can significantly impact the electrical, mechanical, and reliability of subsequent products, and may even render an entire batch of components useless. Therefore, improving the monitoring and quality control capabilities of the cutting process has become a critical link in the quartz device industry chain.

[0003] Traditional quartz cutting quality control relies heavily on manual visual inspection or simple optical imaging methods, making it difficult to achieve high-sensitivity detection of minor defects and early-stage anomalies. As high-end manufacturing demands higher product consistency and yield, more and more companies and research institutions are introducing intelligent sensing and automated detection methods. For example, using high-speed industrial cameras to capture images of cut surfaces, combined with laser spectroscopy technology to analyze changes in the material's surface structure, has become a growing trend in quality monitoring in recent years. At the same time, data-driven analysis methods such as machine vision and artificial intelligence have shown great potential in tasks such as anomaly identification, defect grading, and process tracing in complex process flows. Despite this, further improving the automated identification of cutting anomalies, accurately separating material defects from processing defects, and providing traceable data support for subsequent process optimization remain key technical challenges in the current quartz crystal processing field. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing monitoring method for quartz crystal cutting anomalies is insufficient in identifying tiny defects, has difficulty in distinguishing process anomalies from material background anomalies, and has a limited level of automation.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for monitoring quartz crystal cutting anomalies, comprising:

[0007] The surface image and reflection spectrum of each section during the cutting process are captured by the collaboration of a high-speed industrial camera and a laser scanning spectrometer.

[0008] The surface image and the reflection spectrum are pre-processed and then spatially registered; the spatially registered image is feature-fused with the cutting path of the surface to obtain a clear input data stream;

[0009] Perform pyramid reconstruction on the data stream corresponding to each slice;

[0010] The inherent parameters of the reconstructed section data stream are simplified, and the simplified section data stream is analyzed to obtain the detection result of the quartz crystal section anomaly.

[0011] As a preferred solution of the quartz crystal cutting anomaly monitoring method described in the present invention, the preprocessing includes: performing noise suppression, geometric distortion correction, brightness normalization and artifact removal on the surface image; and performing spectral normalization and interference band removal on the reflection spectrum.

[0012] As a preferred solution of the quartz crystal cutting anomaly monitoring method described in the present invention, the spatial registration includes uniformly mapping the preprocessed data into a standard coordinate system in the physical space, so that the data streams correspond one-to-one in spatial position and achieve pixel-level alignment.

[0013] As a preferred embodiment of the quartz crystal cutting anomaly monitoring method of the present invention, the cutting path includes: using a fixed position of the cutter head as a reference point, using the cutter head lines extending to both sides of the reference point as parameter reflection baselines during cutting, and making the feedback feature vector at the coordinate position of the same parameter reflection baseline equal to the feedback feature vector of the reference point;

[0014] During cutting, the initial position of the reference point, the cutting speed vector and the time are used to record the continuous path of the reference point in the two-dimensional coordinate system where the image is located, and at the same time generate a feedback feature vector reflecting the baseline with the same parameters: ;

[0015] in, represents the 2D pixel coordinates in the spatially registered image, , Represents the path of the reference point in the image; Represents a pixel on the image The feedback feature vector of Cutting speed of the reference point at time t; represents the acceleration of the reference point at time t; represents the normal force of the tool tip at time t; represents the tangential force of the cutter head at time t; represents the cutting trajectory error, which is defined as the Euclidean distance deviation between the actual tool head trajectory and the control instruction; t represents the tth moment in the cutting process;

[0016] Each parameter reflects each pixel in the baseline. A virtual dimension is introduced to indicate that the parameter reflects the actual feedback feature vector of each pixel on the baseline. In this virtual dimension, random assumptions are made about the actual feedback feature vectors of all pixels on both sides.

[0017] The constraints of the random hypothesis are: the sum of the actual feedback feature vectors of all pixels on the parameter-reflected baseline is equal to the feedback feature vector of the parameter-reflected baseline;

[0018] The global fitness of the image is solved, and the random hypothesis combination with a fitness higher than the preset value is selected as the evaluation result; the set of global random hypothesis combinations is obtained ; An represents the nth random hypothesis combination, which is converted into probability in equal proportion according to the fitness corresponding to each element in the set A, so that the set The sum of the probabilities of all elements in is 1;

[0019] The fitness is equal to the sum of the prior probability based on historical data and the rationality analysis result of the one-dimensional simplified convolutional neural network;

[0020] The feature fusion includes: combining random hypotheses corresponding to each element in the set A, and mapping them to the spatially registered images according to the corresponding relationship between the pixels, to obtain n groups of images with increased parameter dimensions.

[0021] As a preferred embodiment of the quartz crystal cutting anomaly monitoring method of the present invention, the pyramid reconstruction comprises: constructing a multi-scale pyramid for each of the n groups of images after parameter dimension increase, performing feature enhancement and contrast learning at each layer, and amplifying the loss of weak anomalies by extracting positive and negative samples;

[0022] By comparing feature representations at different scales and in different regions, the sensitivity of abnormal signals can be enhanced;

[0023] Finally, upsampling and multi-layer fusion reconstruction are performed to obtain the reconstructed image, which is used as the data stream for anomaly detection.

[0024] As a preferred embodiment of the method for monitoring quartz crystal cutting anomalies according to the present invention, the inherent parameters include: abnormal parameters that are identified as having abnormal performance in the data stream after pyramid reconstruction of the standard cutting surface under the same material;

[0025] The simplification process of the inherent parameters includes: performing feature comparison on the inherent parameters in the n groups of acquired data streams, extracting the data components most similar to the inherent parameters from the n groups of data streams, and screening m data streams whose content of similar data components is higher than a preset value as the data streams to be analyzed; in the data streams to be analyzed, the parts that are the same as the abnormal parameters are always represented as regular parameters to obtain the abnormal parts caused only by the cutting abnormality; wherein m≤n.

[0026] As a preferred solution of the quartz crystal cutting anomaly monitoring method according to the present invention, the analysis of the simplified section data stream specifically includes:

[0027] Step 1: Cluster the data features in the m groups of data streams to be analyzed. If there is only one cluster group, proceed to step 3; if there is more than one cluster group, proceed to step 2;

[0028] Step 2: If the ratio of the number of individuals in the group with the most individuals to the number of individuals in the group with the second most individuals is greater than the preset value, only the data stream corresponding to the group with the most individuals is retained and the process goes to step 3; otherwise, the process goes to step 6;

[0029] Step 3: Based on the data stream of each individual in the same cluster family, the abnormal part and the normal part of the image are divided, and the boundary of the division is blurred so that the characteristics of each individual can be expressed in the image after the fuzzy division;

[0030] Step 4: Lock the abnormal core of the m images after fuzzy partitioning and determine the inner and outer boundaries of the fuzzy partitioning;

[0031] Step 5: Use the pre-trained neural network to analyze the pixel features in the inner boundary and obtain the abnormal recognition results;

[0032] Step 6: Calculate the number of individuals in each cluster group according to the number of individuals in the group / the total number of individuals × 100% as the probability coefficient; enter step 3 with the data stream corresponding to each group, and add the corresponding probability coefficient at the output as an additional factor in the evaluation result.

[0033] A monitoring system based on quartz crystal cutting anomalies using the method described in the present invention comprises: an acquisition unit that captures the surface image and reflection spectrum of each cut surface during the cutting process by cooperating with a high-speed industrial camera and a laser scanning spectrometer;

[0034] A processing unit performs spatial registration on the surface image and the reflection spectrum after preprocessing to obtain a registered image; and performs feature fusion on the spatially registered image and the cutting path of the surface to obtain a clear input data stream;

[0035] The optimization unit performs pyramid reconstruction on the data stream corresponding to each slice;

[0036] The detection unit simplifies the inherent parameters of the reconstructed section data stream, analyzes the simplified section data stream, and obtains the detection result of the quartz crystal section abnormality.

[0037] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein: the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0038] A computer-readable storage medium stores a computer program, wherein: when the computer program is executed by a processor, the steps of any one of the methods of the present invention are implemented.

[0039] Beneficial effects of the present invention: The monitoring method based on quartz crystal cutting anomalies provided by the present invention can significantly improve the automation and intelligence level of anomaly detection in the quartz crystal cutting process, and overcome the technical bottleneck of the traditional method's weak recognition ability for tiny defects and low-contrast anomalies. By integrating multimodal data, pyramid multi-scale analysis and inherent parameter assimilation, the precise extraction of multi-granularity and multi-type anomalies on the cutting surface is achieved, and the anomalies caused by the process and the background defects of the material can be effectively distinguished, thereby greatly reducing the risk of misjudgment and missed detection. The method has high-throughput, full-process online detection capabilities, and can support real-time monitoring and quality traceability in large-scale production environments. The proposed technical framework has strong compatibility and is easy to integrate with existing manufacturing and testing equipment. It helps to improve the factory yield, stability and process controllability of quartz components, and provides powerful data and decision-making support for the optimization of quartz processing technology in high-end electronics, optics and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 The first embodiment of the present invention provides an overall flow chart of a method for monitoring quartz crystal cutting anomalies.

[0042] Figure 2 A framework diagram of a monitoring system for quartz crystal cutting anomalies provided in accordance with the second embodiment of the present invention. DETAILED DESCRIPTION

[0043] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0044] Example 1, with reference to Figure 1 , as one embodiment of the present invention, provides a monitoring method based on quartz crystal cutting anomalies, comprising:

[0045] S1: A high-speed industrial camera and a laser scanning spectrometer are used to capture the surface image and reflection spectrum of each section during the cutting process.

[0046] Specifically, high-speed industrial cameras are strategically placed along the cutting path on the cutting table. Selected cameras with high frame rates and high resolution enable real-time imaging of the surface of the cutter's motion area. During the camera acquisition process, LED or laser auxiliary lighting is used to ensure image clarity and the depiction of surface details.

[0047] Simultaneously, a laser scanning spectrometer is deployed next to the cutting area to perform point-by-point or scanning spectral detection on the cutting surface passed by the tool head, and the reflection spectrum curve of the area is obtained in real time.

[0048] The high-speed industrial camera and laser scanning spectrometer are synchronized through an industrial computer or data acquisition system. The corresponding two-dimensional surface image and spectral data of each position are recorded at each cutting moment to ensure the precise correspondence between the image and spectrum in spatial coordinates and cutting timing.

[0049] The obtained raw data provides a high-quality data foundation for subsequent multimodal feature fusion and anomaly detection.

[0050] S2: Pre-processing the surface image and the reflection spectrum and then performing spatial registration; performing feature fusion on the spatially registered image and the cutting path of the surface to obtain a clear input data stream.

[0051] In this embodiment, bilateral filtering is used to suppress noise in surface images; geometric correction based on a calibration plate is used to correct geometric distortion of cross-sectional images; histogram equalization is used to normalize brightness; and area marking based on a strong reflection threshold is used to mark artifacts.

[0052] For reflectance spectra, we used maximum and minimum normalization to normalize spectral signals collected from different batches and under different lighting conditions to a unified range, facilitating data comparison and analysis. We also used a fixed wavelength band shielding method, based on preliminary experiments, to shield specific wavelength ranges where instrument response was abnormal and unrelated to physical properties (such as the multipath artifact region unique to lasers).

[0053] The pre-processed data is uniformly mapped to the standard coordinate system of the physical space, so that the data streams correspond one-to-one in spatial position and pixel-level alignment is achieved.

[0054] The fixed position of the cutter head is used as the reference point, and the cutter head lines extending to both sides of the reference point are used as the parameter reflection baseline during cutting, so that the feedback feature vector at the coordinate position of the same parameter reflection baseline is equal to the feedback feature vector of the reference point.

[0055] During cutting, the initial position of the reference point, the cutting speed vector and the time are used to record the continuous path of the reference point in the two-dimensional coordinate system where the image is located during the cutting process.

[0056] The tool head reference point path is parameterized, and the initial position of the tool head reference point is defined as: .

[0057] Assume the cutting velocity vector is: .

[0058] The reference point on the cutting path is at any time The two-dimensional coordinates of are:

[0059] ;

[0060] In practical engineering, the trajectory can be recorded by discrete integration, sensor sampling or motion control feedback. Indicates the horizontal coordinate of the initial position of the tool head reference point, Indicates the ordinate of the initial position of the tool head reference point, Indicates that the tool head reference point is The speed in the x-axis direction at the moment; Indicates that the tool head reference point is The speed in the y-axis direction at the moment; Indicates the index of the timestamp. t indicates the current moment.

[0061] At the same time, the feedback feature vector reflecting the same parameter baseline is generated: ;

[0062] in, represents the 2D pixel coordinates in the spatially registered image, , Represents the path of the reference point in the image; Represents a pixel on the image The feedback feature vector of Cutting speed of the reference point at time t; represents the acceleration of the reference point at time t; represents the normal force of the tool tip at time t; represents the tangential force of the cutter head at time t; Represents the cutting trajectory error, which is defined as the Euclidean distance deviation between the actual tool head trajectory and the control instruction.

[0063] When each parameter reflects each pixel in the baseline, a virtual dimension is introduced to indicate that the parameter reflects the actual feedback feature vector of each pixel on the baseline; on the virtual dimension, random assumptions are made on the actual feedback feature vectors of all pixels on both sides.

[0064] The constraint of the random hypothesis is that the sum of the actual feedback feature vectors of all pixels on the parameter-reflecting baseline is equal to the feedback feature vector of the parameter-reflecting baseline.

[0065] The global fitness of the image is solved, and the random hypothesis combination with a fitness higher than the preset value is selected as the evaluation result; the set of global random hypothesis combinations is obtained ; A1, A2, An represent the first, second, and nth random hypothesis combinations respectively. According to the fitness corresponding to each element in the set A, they are converted into probabilities in equal proportion, so that the set The sum of the probabilities of all elements in is 1.

[0066] What needs to be said is that by taking the fixed reference point of the cutter head as the origin of the cutting parameters, combined with the real-time recorded cutting velocity vector and time information, the continuous trajectory of the cutter head reference point in two-dimensional space is accurately parameterized. Furthermore, along the normal direction of the reference point path (i.e., the parameter reflection baseline), the physical feedback parameters of the reference point (such as velocity, acceleration, normal force, tangential force, trajectory error, etc.) are assigned to all pixels on the baseline, realizing the mapping and expansion of physical information in space. In this way, all pixel points in the same parameter reflection baseline are regarded as homogeneous units in the feedback feature space, uniformly expressing the physical state of the cutting process.

[0067] To improve the ability to represent complex, real-world anomalies, this method introduces a virtual dimension for each pixel within each parameter-reflecting baseline. It then generates several sets of random hypotheses for all actual feedback feature vectors, with the constraint that the sum of all pixel feedback feature vectors under each set of random hypotheses must equal the baseline's overall feedback feature vector. This approach enables the system to "regularize" high-dimensional modeling and inference of irregular, non-uniform, unknown noise, and multi-source anomalies, combining two-dimensional space with a virtual random space.

[0068] Globally, through fitness calculations, highly credible hypothesis combinations are selected as valid evaluation results, and the probabilities of all random hypothesis combinations are normalized, ensuring that the analysis results are both statistically reliable and physically interpretable. Overall, this solution effectively introduces the third dimension of "structured randomness," breaking through the limitations of describing heterogeneous and complex anomalies in two-dimensional space, and laying a solid data and mathematical model foundation for intelligent detection of cutting anomalies and process traceability.

[0069] The fitness is equal to the sum of the prior probability based on historical data and the rationality analysis results of the one-dimensional simplified convolutional neural network. Among them, the one-dimensional simplified convolutional neural network is used to determine whether each set of hypotheses (i.e., the distribution of physical parameters of each pixel) is possible, physically reasonable, and has no internal contradictions in "structural mechanics / material mechanics". Core: The distribution of physical quantities (such as normal force, tangential force, stress, displacement, velocity, etc.) must conform to the physical laws of force balance, continuity, stress-strain relationship, etc. under real materials and structures. The structure of the one-dimensional simplified convolutional neural network (1D-CNN) is as follows:

[0070] Input layer: The global image after each random hypothesis combination configuration, where each pixel has multi-dimensional physical features.

[0071] Convolutional layer (1D Conv): The receptive field of the convolution kernel can cover a local physical area, which is equivalent to sliding window mechanical analysis.

[0072] Pooling layer: aggregates local features to simulate stress continuity and overall trends.

[0073] Fully connected layer (FC): global mechanical discrimination.

[0074] Output layer: Sigmoid activation, output rationality probability score.

[0075] The feature fusion includes: combining random hypotheses corresponding to each element in the set A, and mapping them to the spatially registered images according to the corresponding relationship between the pixels, to obtain n groups of images with increased parameter dimensions.

[0076] By taking as input the distribution of physical quantities (such as normal force, tangential force, stress, and velocity) on the baseline as parameters for each set of random hypotheses, a one-dimensional simplified convolutional neural network (1D-CNN) was designed to automatically determine the plausibility of the hypotheses at the physical and structural mechanics levels. The 1D-CNN's structure enables end-to-end modeling and characterization of physical quantities in space, mechanical equilibrium, and stress-strain relationships, effectively identifying hypotheses that contain internal physical contradictions, sudden local stress changes, or violate the laws of material mechanics.

[0077] This approach not only considers prior probabilities derived from historical data statistics but also incorporates complex mechanical rationality as a key element in fitness assessment, significantly enhancing the physical realism and engineering reliability of anomaly detection. Hypotheses with high fitness represent parameter distributions that are both statistically supported and mechanically reasonable, and are more consistent with defects and anomalies in real-world engineering scenarios.

[0078] During the feature fusion phase, all highly adaptive hypothesis combinations are used as parameter mapping templates and embedded into the spatially registered original image data stream, in a one-to-one correspondence at the pixel level. This generates multiple sets of new data streams with increasing parameter dimensions. These new data streams retain the original spatial and optical information while superimposing physical feedback and rationality judgment dimensions, providing a rich, multi-layered data foundation for subsequent pyramid reconstruction and intelligent detection. Ultimately, the system is able to achieve comprehensive, high-precision identification of complex cutting anomalies and trace their physical mechanisms, ensuring the scientific nature, accuracy, and interpretability of the analysis results.

[0079] S3: Perform pyramid reconstruction on the data stream corresponding to each slice.

[0080] For n groups of images after the parameter dimension is increased, perform the following operations respectively:

[0081] Multi-scale pyramid construction: high-dimensional input feature tensor after spatial registration , through multi-level downsampling, sequentially generate Pyramid features with decreasing layer resolution:

[0082] ;

[0083] Among them, h is the native resolution. represents the high-dimensional input feature tensor after spatial registration, represents the original horizontal axis coordinate, Represents the original vertical axis coordinate. Indicates the Layer pyramid feature tensor, with decreasing resolution, Represents the h-1th layer pyramid feature tensor; Indicates the Layer downsampling operators, such as Gaussian downsampling, strided convolution, etc. Indicates the The horizontal and vertical coordinates corresponding to the layer; Indicates the current layer number of the pyramid, from 0 to . Indicates the total number of pyramid layers.

[0084] Feature enhancement and contrast learning structure at each layer: In each pyramid layer, independent feature extraction and weak anomaly amplification modules are used. The contrast learning structure and custom loss mechanism are introduced to automatically amplify low-contrast defect signals:

[0085] First, feature representations (positive samples and negative samples) are extracted for normal areas and weak abnormal areas at each scale:

[0086] ;

[0087] Construct a self-supervised contrast loss (such as InfoNCE and Triplet loss) to bring similar features closer together and distance features of different classes (normal and abnormal), especially increasing the weight of low-amplitude abnormal signals.

[0088] Customized weak anomaly amplification loss at each layer:

[0089] ;

[0090] in, It represents the weak anomaly amplification loss of the hth layer, which is used to enhance the low-contrast defect signal. represents the abnormal amplification weight hyperparameter of the h-th layer. Indicates the number of sampling points in the h-th layer of weak anomaly area. Indicates the current sampling point number, ranging from 1 to . The feature vector representing the sth weak abnormal region in the hth layer. Represents the reference feature mean vector of the normal region (positive sample) of the h-th layer.

[0091] Multi-scale contrastive loss:

[0092] By comparing feature representations at different scales and in different regions, the model’s sensitivity to abnormal signals is enhanced:

[0093] ;

[0094] in, represents the h-th layer multi-scale contrast loss. Indicates the number of sampling points used for comparison in the hth layer. Indicates the current sample number in the contrast loss, ranging from 1 to . Indicates the hth layer The feature vector of samples (positive samples). Represents the feature mean vector (negative sample) of the weak abnormal area in the hth layer. represents the temperature scaling parameter of the hth layer. Represents the numbers of all comparison samples in the denominator. The feature vector representing the kth sample at the hth level of the pyramid is the feature expression of the positive sample or all samples used for multi-scale contrast loss calculation.

[0095] Upsampling and multi-layer fusion reconstruction:

[0096] The enhanced and contrast-optimized features of each layer are uniformly regressed to the original resolution through upsampling and fused to form the final output feature tensor:

[0097] ;

[0098] in, Represents the high-dimensional feature tensor of the final fusion output, with the resolution of the original input size. ) represents multi-layer feature fusion operations (such as weighting, splicing, attention mechanism, etc.). US Indicates the Layer upsampling operator, used to restore the feature tensor to the original resolution. Indicates the The feature tensor of the layer after feature enhancement and contrast optimization. Indicates the pyramid layer number, ranging from 0 to .

[0099] The total loss is the weighted sum of the weak anomaly amplification loss and the multi-scale contrast loss at each layer, as well as the global detection or segmentation loss:

[0100] ;

[0101] in, represents the overall training loss function. Indicates the Weak anomaly amplification loss of the layer. Indicates the Weighting factor for layer contrastive loss. Indicates the Multi-scale contrastive loss for layers. Represents the weight hyperparameter of the global detection or segmentation loss. Represents the global detection or segmentation loss of the final output.

[0102] Finally, upsampling and multi-layer fusion reconstruction are performed to obtain the reconstructed image, which is used as the data stream for anomaly detection.

[0103] It is important to note that the multi-scale feature pyramid structure enables comprehensive, multi-layered anomaly information capture and intelligent enhancement of high-dimensional input data streams after the cutting surface parameters have been increased in dimensionality. Through multi-level downsampling, the original spatially registered feature tensor is gradually decomposed into pyramid feature layers of varying resolutions. Each layer focuses on structural features at a different spatial scale, thus addressing the detection needs of both large-scale structural anomalies and small, weak signal anomalies.

[0104] At each pyramid level, independent feature enhancement and anomaly amplification modules are implemented. A self-supervised contrastive learning mechanism and a custom anomaly loss function are explicitly introduced, enabling the model to proactively distinguish features from normal and weakly abnormal regions, while also amplifying and enhancing the significance of low-contrast, weak anomalies. Contrastive loss further increases the distribution distance between normal and abnormal features, effectively improving the model's sensitivity and robustness to complex anomalies.

[0105] The feature outputs of all pyramid layers are upsampled and multi-level fused to restore them to their original resolution, forming a high-dimensional output tensor containing multi-scale spatial information and anomaly-enhanced features, providing a multi-angle, information-dense input foundation for downstream anomaly detection. Ultimately, through global loss integration, layer-by-layer amplification and global high-precision recognition of weak anomaly signals are achieved, overcoming the technical bottlenecks of traditional segmentation / detection models, which have limited low-contrast anomaly recognition capabilities and spatial scale, and significantly improving the detection sensitivity and engineering practicality of quartz crystal cutting anomalies.

[0106] S4: Simplifying the inherent parameters of the reconstructed section data stream, analyzing the simplified section data stream, and obtaining the detection result of the quartz crystal section anomaly.

[0107] The inherent parameters include: abnormal parameters that have been identified as abnormal in a data stream after pyramid reconstruction of a standard cutting surface under the same material. The inherent parameter simplification process includes: comparing the characteristics of the inherent parameters in n acquired data streams, extracting the data components most similar to the inherent parameters from the n data streams, and selecting m data streams with similar data components exceeding a preset value as data streams to be analyzed; in the data streams to be analyzed, the parts that are consistent with the abnormal parameters are always represented as normal parameters, thereby obtaining the abnormal parts caused only by the cutting abnormality; where m≤n.

[0108] In this solution, by introducing a third random dimension, the feedback feature vectors of each parameter of the cutting surface reflecting the pixel points on the baseline are decomposed into multiple sets of hypotheses, achieving the expression and regularization of complex anomaly distributions in high-dimensional space. This mechanism greatly enhances the system's ability to capture diverse anomalies, but it also brings a potential risk: in the hypothesis space, some hypothesis combinations may introduce too much freedom, resulting in over-correction of the original physical characteristics. In other words, some data streams may be "corrected" to deviate from the actual material or process background characteristics, resulting in weakened physical interpretability and even loss of perception of the actual nature of the anomaly.

[0109] Therefore, it's essential to proactively screen all generated data streams for those that closely resemble known material parameters. This ensures that, even after introducing random assumptions and high-dimensional mapping, the selected data streams retain the most consistent anomaly characteristics with the true material background, thus avoiding algorithmic drift and physical distortion caused by the "third dimension." Only these closest data streams are physically plausible representations of the objective distribution of material anomalies. Subsequent analysis can be based on the true process environment and material background, ensuring scientific and reliable testing.

[0110] In addition, by setting "similarity content higher than the preset value" as the screening criterion, the intensity of inherent parameter retention can be flexibly adjusted, the sensitivity of the algorithm and physical constraints can be balanced, and the generalization of anomaly detection can be effectively prevented from getting out of control and the misjudgment rate from increasing.

[0111] Furthermore, the simplified section data flow is analyzed, specifically including:

[0112] Step 1: Cluster the data features of the m data streams to be analyzed. If only one cluster exists, proceed to Step 3; if more than one cluster exists, proceed to Step 2. By clustering the features of multiple data streams, abnormal expression patterns can be automatically classified. This design aims to reveal the inherent structural relationships between data, provide a hierarchical basis for subsequent anomaly identification, and reduce bias caused by subjective classification.

[0113] Step 2: If the ratio of the number of individuals in the cluster with the most individuals to the number of individuals in the cluster with the second most individuals is greater than a preset value, only the data stream corresponding to the cluster with the most individuals is retained, and the process proceeds to Step 3; otherwise, the process proceeds to Step 6. (Normal cases proceed to Step 3 because the characteristic performance of the anomaly is minimal and can be equivalent through fuzzy concepts. However, the content of Step 6 is an anomaly, and in this case, the regional differences in the anomaly performance may be quite significant, so it cannot be simply replaced through fuzzy concepts.) By establishing a judgment threshold for the ratio of the number of individuals between clusters, the processing strategy can be dynamically adjusted in scenarios with clear commonalities and significant differences. This diversion mechanism avoids excessive blurring caused by a small number of outliers, allowing the mainstream features to be more powerfully expressed.

[0114] Step 3: Based on the data stream of each individual in the same cluster family, the abnormal part and the normal part of the image are divided, and the boundary of the division is blurred (in this case, it is basically caused by slight differences in the boundary of the abnormal part. Therefore, by blurring the abnormal boundary, it can include the characteristics of all individuals), so that the characteristics of each individual can be expressed in the blurred divided image.

[0115] It is important to note that, in situations where there are slight differences in the boundaries of anomaly features within multiple high-dimensional data streams, fuzzification is introduced to moderately fuse the boundary information between individuals, making the clustering results more inclusive and representative when expressing anomaly regions. This fuzzification process can effectively alleviate boundary uncertainty caused by noise, local perturbations, or microstructural differences, thereby more accurately identifying and targeting core anomaly regions in the data stream. Furthermore, when there are significant differences between clusters, independent analysis of each cluster and assigning probability weights can account for the diversity of different anomaly patterns, making subsequent anomaly identification both robust and uncertainty-retaining, thereby enhancing the scientific nature and engineering adaptability of the overall analysis.

[0116] Although the concept of fuzzification is introduced in this embodiment, this solution actually simplifies the fuzzy concept. After blurring the boundaries of the region divisions, only the middle portion with consistent data is needed. Therefore, the blurred boundary band is divided into inner and outer boundaries, and only the inner boundary is used to extract the "middle portion with consistent data." This simplification also simplifies the algorithm implementation: the original fuzzy concept required superimposing the anomaly masks of each individual within the same cluster by pixel position to obtain a probability map, and then calculating the confidence level based on the probability to achieve fuzzification. In this solution, by performing boundary identification on the abnormal region of each individual and superimposing the identification results, an abnormal boundary identification result is ultimately obtained by superimposing several data streams. Then, within this boundary identification result, the inner and outer boundaries of the separation band are found for each abnormal region. This simplifies the identification purpose of fuzzification.

[0117] Step 4: Lock the abnormal core of the m images after fuzzy partitioning and determine the inner and outer boundaries of the fuzzy partitioning.

[0118] Step 5: Use the pre-trained neural network to analyze the pixel features in the inner boundary and obtain the abnormal recognition results.

[0119] Step 6: Calculate the number of individuals in each cluster, using the probability coefficient (number of individuals in the cluster / total number of individuals) × 100. Enter the data stream corresponding to each cluster into Step 3, and add the corresponding probability coefficient to the output as an additional factor in the evaluation result. By assigning probability weights to the results of different clusters, the final output is both subjectively robust and diverse in its expression, reflecting the uncertainty of the data itself and facilitating risk quantification and traceability for subsequent decision-making.

[0120] In this embodiment, the neural network may be a convolutional neural network (CNN), such as U-Net, ResNet, DenseNet, or EfficientNet. Convolutional neural networks have powerful automatic extraction capabilities for spatial structure and local texture features, making them suitable for processing spatially registered image data streams, pyramid features, and fine-grained identification of abnormal regions. CNNs can leverage pre-trained models for transfer learning, significantly improving detection performance and generalization capabilities in scenarios with limited data or complex features. Modern CNNs support end-to-end input and output, facilitating integration with the entire automated detection process, and enabling efficient, real-time localization, segmentation, and classification of core anomalies. In other embodiments, the neural network may also be an attention-based network, a graph neural network (GNN), a recurrent neural network (RNN / LSTM / GRU), a generative adversarial network (GAN), or the like.

[0121] Example 2, as Figure 2 As shown, this embodiment also provides a monitoring system based on quartz crystal cutting anomalies, which includes:

[0122] The acquisition unit uses a high-speed industrial camera and a laser scanning spectrometer to capture the surface image and reflection spectrum of each section during the cutting process.

[0123] The processing unit performs spatial registration after pre-processing on the surface image and the reflection spectrum; and performs feature fusion on the spatially registered image and the cutting path of the surface to obtain a clear input data stream.

[0124] The optimization unit performs pyramid reconstruction on the data stream corresponding to each slice.

[0125] The detection unit simplifies the inherent parameters of the reconstructed section data stream, analyzes the simplified section data stream, and obtains the detection result of the quartz crystal section abnormality.

[0126] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0127] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0128] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, and then editing, interpreting, or processing in another suitable manner as necessary, and then storing it in a computer memory.

[0129] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the aforementioned embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or combination of the following technologies known in the art may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A monitoring method based on quartz crystal cutting anomaly, characterized in that: include: The surface image and reflection spectrum of each section during the cutting process are captured by the collaboration of a high-speed industrial camera and a laser scanning spectrometer. The surface image and the reflection spectrum are pre-processed and then spatially registered to obtain a registered image; the spatially registered image is feature-fused with the cutting path of the surface to obtain a clear input data stream; Perform pyramid reconstruction on the data stream corresponding to each slice; The inherent parameters of the reconstructed cross-sectional data stream are simplified, and the simplified cross-sectional data stream is analyzed to obtain the detection result of the abnormality of the cross-sectional surface of the quartz crystal; The cutting path includes: using a fixed position of the cutter head as a reference point, using the cutter head lines extending to both sides of the reference point as parameter reflection baselines during cutting, and making the feedback feature vector at the coordinate position of the same parameter reflection baseline equal to the feedback feature vector of the reference point; During cutting, the initial position of the reference point, the cutting speed vector and the time are used to record the continuous path of the reference point in the two-dimensional coordinate system where the image is located, and at the same time generate a feedback feature vector reflecting the baseline with the same parameters: ； in, represents the 2D pixel coordinates in the spatially registered image, , Represents the path of the reference point in the image; Represents a pixel on the image The feedback feature vector of Cutting speed of the reference point at time t; represents the acceleration of the reference point at time t; represents the normal force of the tool tip at time t; represents the tangential force of the cutter head at time t; represents the cutting trajectory error; t represents the t-th moment in the cutting process; Each parameter reflects each pixel in the baseline. A virtual dimension is introduced to indicate that the parameter reflects the actual feedback feature vector of each pixel on the baseline. In this virtual dimension, random assumptions are made about the actual feedback feature vectors of all pixels on both sides. The constraints of the random hypothesis are: the sum of the actual feedback feature vectors of all pixels on the parameter-reflected baseline is equal to the feedback feature vector of the parameter-reflected baseline; The global fitness of the image is solved, and the random hypothesis combination with a fitness higher than the preset value is selected as the evaluation result; the set of global random hypothesis combinations is obtained ; An represents the nth random hypothesis combination, which is converted into probability in proportion to the fitness corresponding to each element in set A; The fitness is equal to the sum of the prior probability based on historical data and the rationality analysis result of the one-dimensional simplified convolutional neural network; The feature fusion comprises: combining the random hypotheses corresponding to each element in the set A, and mapping them to the spatially registered images according to the corresponding relationship of the pixels, to obtain n groups of images with increased parameter dimensions; The inherent parameters include: abnormal parameters that are identified as having abnormal performance in the data stream after pyramid reconstruction of the standard cutting surface under the same material; The inherent parameter simplification process includes: performing feature comparison on the inherent parameters in n acquired data streams, extracting data components most similar to the inherent parameters from the n data streams, selecting m data streams with similar data component contents higher than a preset value as data streams to be analyzed; in the data streams to be analyzed, representing the portion identical to the abnormal parameter as a normal parameter, thereby obtaining the abnormal portion caused only by the cutting abnormality; Where m≤n.

2. The method for monitoring quartz crystal cutting anomalies according to claim 1, wherein: The preprocessing includes: performing noise suppression on the surface image, geometric distortion correction on the section image, brightness normalization and artifact removal; and performing spectrum normalization and interference band removal on the reflection spectrum.

3. The method for monitoring quartz crystal cutting anomalies according to claim 2, wherein: The spatial registration includes: uniformly mapping the pre-processed data into a standard coordinate system of the physical space, so that the data streams correspond one to one in spatial position and achieve pixel-level alignment.

4. The method for monitoring quartz crystal cutting anomalies according to claim 3, wherein: The pyramid reconstruction includes: for n groups of images after parameter dimension increase, respectively performing: multi-scale pyramid construction, respectively performing feature enhancement and contrast learning at each layer, and amplifying the loss of weak anomalies by extracting positive samples and negative samples; By comparing feature representations at different scales and in different regions, the sensitivity of abnormal signals can be enhanced; Finally, upsampling and multi-layer fusion reconstruction are performed to obtain the reconstructed image, which is used as the data stream for anomaly detection.

5. The method for monitoring quartz crystal cutting anomalies according to claim 4, wherein: The analysis of the simplified section data flow specifically includes: Step 1: Cluster the data features in the m groups of data streams to be analyzed. If there is only one cluster group, proceed to step 3; if there is more than one cluster group, proceed to step 2; Step 2: If the ratio of the number of individuals in the group with the most individuals to the number of individuals in the group with the second most individuals is greater than the preset value, only the data stream corresponding to the group with the most individuals is retained and the process goes to step 3; otherwise, the process goes to step 6; Step 3: Based on the data stream of each individual in the same cluster family, the abnormal part and the normal part of the image are divided, and the boundary of the division is blurred so that the characteristics of each individual can be expressed in the image after the fuzzy division; Step 4: Lock the abnormal core of the m images after fuzzy partitioning and determine the inner and outer boundaries of the fuzzy partitioning; Step 5: Use the pre-trained neural network to analyze the pixel features in the inner boundary and obtain the abnormal recognition results; Step 6: Calculate the number of individuals in each cluster group according to the number of individuals in the group / the total number of individuals × 100% as the probability coefficient; enter step 3 with the data stream corresponding to each group, and add the corresponding probability coefficient at the output as an additional factor in the evaluation result.

6. A quartz crystal cutting anomaly monitoring system using the method according to any one of claims 1 to 5, characterized in that: It includes an acquisition unit that uses a high-speed industrial camera and a laser scanning spectrometer to capture the surface image and reflection spectrum of each section during the cutting process; a processing unit, performing spatial registration after pre-processing on the surface image and the reflection spectrum to obtain a registered image; The spatially registered image is feature-fused with the surface cutting path to obtain a clear input data stream; The optimization unit performs pyramid reconstruction on the data stream corresponding to each slice; The detection unit simplifies the inherent parameters of the reconstructed section data stream, analyzes the simplified section data stream, and obtains the detection result of the quartz crystal section abnormality.

7. A computer device comprising: memory and processor; The memory stores a computer program, wherein the processor implements the steps of the method according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images