Control method of quartz crystal resonator

Abstract

The invention provides a control method of a quartz crystal resonator, and relates to the technical field of resonator control, and the method comprises the steps: obtaining a multi-dimensional resonance performance parameter sequence of the quartz crystal resonator; constructing a dynamic resonance behavior prediction model by using the multi-dimensional resonance performance parameter sequence; combining with the dynamic resonance behavior prediction model, identifying the operation state abnormity of the quartz crystal resonator, and generating a fault prediction result; and performing fault type judgment and diagnosis output according to a fault prediction result, and generating a cooperative control strategy to execute distributed control. According to the invention, the technical problem that the state change and the potential fault trend of the resonator in a complex environment cannot be accurately identified in the prior art can be solved, and the technical target of intelligent adjustment of the operation state of the quartz crystal resonator based on dynamic prediction and cooperative control is achieved. The technical effects of improving the control precision, the adaptive capability and the overall reliability of the system are achieved.

Description

Technical Field

[0001] The present application relates to the field of resonator control technology, and in particular to a control method for a quartz crystal resonator. Background Art

[0002] Quartz crystal resonators, as high-precision frequency control components, are widely used in communications, navigation, industrial control, aerospace and other fields. Their stability and reliability have a significant impact on the operating performance of the entire system.

[0003] Currently, existing technologies primarily employ control logic based on single-dimensional signal feedback, such as temperature compensation via single-point temperature sampling or fixed drive regulation strategies triggered by frequency offsets. This static, rule-driven control approach lacks the ability to model the complex coupling relationships between multiple sources, making it ineffective in predicting and responding to the evolving resonant behavior under dynamic conditions. This leads to delayed fault response, insufficient regulation accuracy, and low overall system control efficiency.

[0004] In summary, the existing technology has technical problems such as the lack of multi-dimensional perception fusion and dynamic behavior modeling capabilities, which leads to the control system being unable to accurately identify the state changes and potential failure trends of the resonator in complex environments, further affecting the stable operation and precise control of the quartz crystal resonator in high-dynamic application scenarios. Summary of the Invention

[0005] The purpose of this application is to provide a control method for a quartz crystal resonator to solve the technical problem in the existing technology that, due to the lack of multi-dimensional perception fusion and dynamic behavior modeling capabilities, the control system is unable to accurately identify the state changes and potential failure trends of the resonator in a complex environment, further affecting the stable operation and precise control of the quartz crystal resonator in high-dynamic application scenarios.

[0006] In view of the above problems, the present application provides a control method for a quartz crystal resonator, including: obtaining a multi-dimensional resonance performance parameter sequence of the quartz crystal resonator; using the multi-dimensional resonance performance parameter sequence to construct a dynamic resonance behavior prediction model; combining the dynamic resonance behavior prediction model to identify abnormal operating status of the quartz crystal resonator and generate a fault prediction result; performing fault type judgment and diagnostic output based on the fault prediction result, and generating a collaborative control strategy to execute distributed control.

[0007] Preferably, the control method of a quartz crystal resonator also includes: performing time series alignment and noise filtering on the multi-dimensional resonance performance parameter sequence to obtain an initial preprocessing sequence; normalizing and standardizing the initial preprocessing sequence according to preset sampling rules to obtain a preprocessing sequence; and distributing the preprocessing sequence to an edge computing device.

[0008] Preferably, the control method of a quartz crystal resonator also includes: extracting parameter characteristics and resonance characteristics of the preprocessing sequence based on the edge computing device, and inputting them into the neural network structure for training; generating a dynamic resonance behavior prediction model through the training results; performing control data prediction on real-time preprocessing data according to the dynamic resonance behavior prediction model, and performing performance evaluation; generating compensation parameters based on the performance evaluation results and integrating them into the control unit.

[0009] Preferably, the control method of a quartz crystal resonator also includes: extracting sensitive factors from the preprocessing sequence based on the performance evaluation results, and establishing a dynamically optimized sensitive mapping relationship between parameters and performance indicators; based on the real-time preprocessing data, fine-tuning the parameter characteristic data in the sensitive mapping relationship, and sending the optimized control parameters to the control unit in real time; monitoring the optimized control effect and performing feedback correction.

[0010] Preferably, the control method of a quartz crystal resonator further includes: extracting the residuals between real-time sensing data and predicted control data, inputting the residual sequence into a fault identifier to identify the fault mode corresponding to the absolute value of the residual in the residual sequence; and outputting a fault prediction result according to the fault mode.

[0011] Preferably, the control method of a quartz crystal resonator also includes: introducing a residual connection structure according to the residual sequence, and designing a recursive feedback module in combination with time series information; training the response sensitivity of the sensitive factor through an attention mechanism; and using continuous training results to adjust the parameter feature data weight distribution of the dynamic resonance behavior prediction model in the recursive feedback module.

[0012] Preferably, the control method of a quartz crystal resonator also includes: retrieving fault sample data; performing a similarity comparison between the fault mode and the fault sample data; generating a preliminary diagnostic conclusion based on the similarity results; calling a diagnostic mechanism to determine the fault development trend, generating a diagnostic output result including the fault type, warning level, possible location and recommended response, and adding it to the fault prediction result.

[0013] Preferably, the control method of a quartz crystal resonator further includes: adaptively updating the control model parameters of the dynamic resonance behavior prediction model at a learning rate according to the residual sequence; and saving a model snapshot and verifying its effectiveness during the evolution of the collaborative control strategy.

[0014] Preferably, the control method of a quartz crystal resonator also includes: initializing the information communication link between multiple resonator nodes, and collecting the operating data of each node in real time and performing standardized processing; after standardized processing, determining the collaborative control strategy according to the operating status of each node and executing distributed control; dynamically switching the collaborative control priority according to the fault prediction result.

[0015] Preferably, the control method of a quartz crystal resonator also includes: identifying the trigger conditions for the decrease in control accuracy, and performing strategy correction based on the trigger conditions; performing online verification of the strategy correction results and retaining the optimal version; constructing an evolutionary map based on the historical optimization path; and performing multi-strategy fusion based on the evolutionary map to generate an optimized control strategy.

[0016] The technical solution provided in this application has at least the following technical effects or advantages: by realizing the technical goal of intelligent adjustment of the operating state of the quartz crystal resonator based on dynamic prediction and collaborative control, the technical effect of improving control accuracy, adaptability and overall system reliability is achieved.

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, which can be implemented in accordance with the contents of the description, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically listed below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in this application or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and a person of ordinary skill in the art can obtain other drawings based on the provided drawings without creative work.

[0019] Figure 1 This is a flow chart of a control method for a quartz crystal resonator according to the present application.

[0020] Figure 2 This is a flow chart of parameter preprocessing in a control method for a quartz crystal resonator of the present application. DETAILED DESCRIPTION

[0021] This application provides a control method for a quartz crystal resonator, resolving the existing technical problem that, due to a lack of multi-dimensional perception fusion and dynamic behavior modeling capabilities, the control system cannot accurately identify the state changes and potential failure trends of the resonator in complex environments, further affecting the stable operation and precise control of the quartz crystal resonator in highly dynamic application scenarios. This application achieves the technical goal of intelligently adjusting the operating state of the quartz crystal resonator based on dynamic prediction and collaborative control, thereby achieving the technical effect of improving control accuracy, adaptability, and overall system reliability.

[0022] Below, the technical solutions in this application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this application, rather than all the embodiments of this application. It should be understood that this application is not limited to the example embodiments described herein. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. It should also be noted that, for the convenience of description, only the parts related to this application, rather than all of them, are shown in the accompanying drawings.

[0023] For example 1, please refer to the attached Figure 1 , the present application provides a control method for a quartz crystal resonator, which specifically includes the following steps:

[0024] S1: Obtain a multi-dimensional resonance performance parameter sequence of a quartz crystal resonator.

[0025] Specifically, obtaining a multi-dimensional resonant performance parameter sequence for a quartz crystal resonator involves real-time monitoring of multiple key performance dimensions during operation using various types of sensors deployed around or within the resonator, and arranging the continuously collected data in chronological order to form a sequence. Multi-dimensionality typically refers to multiple parameter dimensions, including physical quantities such as resonant frequency, amplitude, circuit drive voltage, temperature, pressure, and quality factor that reflect the device's operating state. Performance parameters are quantitative descriptions of these physical quantities, such as a frequency of 10 MHz, a temperature of 35 degrees Celsius, and an amplitude of 2 volts at a certain moment. A sequence is a collection of data that changes continuously at multiple time points, facilitating subsequent modeling, prediction, and control optimization. For example, with a sampling frequency of once per second, sixty sets of parameter values ​​can be obtained within one minute, each containing data for dimensions such as frequency, amplitude, and temperature, forming a multi-dimensional parameter sequence of sixty time steps.

[0026] S2: Construct a dynamic resonance behavior prediction model using a multi-dimensional resonance performance parameter sequence.

[0027] Specifically, using a multi-dimensional resonant performance parameter sequence to construct a dynamic resonant behavior prediction model means taking continuous data containing multiple performance indicators at multiple time points as input, and using machine learning or deep learning methods to establish a mathematical model that can reflect the changing trend of the resonator's behavior. The multi-dimensional resonant performance parameter sequence is a multi-dimensional data set composed of multiple parameters such as resonant frequency, amplitude, voltage, temperature, etc. in chronological order, which can reflect the comprehensive operating characteristics of the resonator under different environments and working conditions; dynamic means that the model is not only applicable to the current state, but can also be adaptively updated over time to predict possible future state changes; resonant behavior refers to the vibration characteristics of a quartz crystal under specific conditions, including key phenomena such as frequency fluctuations and power consumption changes; the prediction model is a functional expression structure that can estimate the operating parameters at a certain moment in the future after learning the relationship between input data and output results through an algorithm.

[0028] S3: In combination with the dynamic resonance behavior prediction model, identify abnormal operating status of the quartz crystal resonator and generate a fault prediction result.

[0029] Specifically, a predictive model constructed based on a multi-dimensional resonant performance parameter sequence is used to analyze the real-time operating data of the quartz crystal resonator. Abnormal behavior is determined by comparing the deviation between the model's predicted values ​​and the actual observed values. The dynamic resonant behavior prediction model can capture nonlinear relationships between parameters based on historical data, such as the effect of temperature changes on frequency, thereby generating expectations for future vibration performance when the latest status data is input. An abnormal operating state refers to a significant deviation between the resonator's current actual output data and the model's predicted results, manifested as frequency jumps, sudden drops in amplitude, and sudden increases in power consumption. Deviations outside the normal noise fluctuation range indicate that the device may be in a degraded or pre-failure state. Fault prediction results are generated by performing a comprehensive evaluation using a diagnostic model after detecting such deviations, generating inferred information about the fault type, probability of occurrence, and possible impact.

[0030] S4: Perform fault type judgment and diagnosis output based on the fault prediction result, generate a collaborative control strategy and execute distributed control.

[0031] Specifically, after obtaining predicted potential fault trends, the system further analyzes the specific fault type and outputs structured diagnostic information. This information is then used to develop targeted control strategies, coordinating multiple resonator nodes to respond to abnormal situations in a distributed manner. The fault prediction results include identification information for features such as frequency anomalies, temperature drift, and voltage imbalance. By comparing these features with existing fault patterns in a historical sample library, the system determines the fault's category, such as thermal compensation failure or drive instability. After determining the fault type, the system outputs the analysis results in a report containing the fault name, possible location, severity level, and recommended actions, which can be used to support automated control systems or manual intervention. A coordinated control strategy dynamically allocates tasks among multiple resonator nodes based on their respective states, maintaining overall system stability through information sharing and load shifting. Distributed control means that rather than centralized management at a single point, each node independently completes regulatory actions and synchronously responds to global policies, thereby improving system responsiveness and fault tolerance.

[0032] Further, if Figure 2 As shown, the present application also includes: performing time series alignment and noise filtering on the multi-dimensional resonance performance parameter sequence to obtain an initial preprocessing sequence; normalizing and standardizing the initial preprocessing sequence according to preset sampling rules to obtain a preprocessing sequence; and distributing the preprocessing sequence to an edge computing device.

[0033] Specifically, the multi-dimensional resonance performance parameter series is time-series aligned and noise filtered. This means that multiple acquired parameter data reflecting the resonator state, such as frequency, amplitude, temperature, and pressure, are aligned based on the acquisition time, placing them on the same time base for easier processing and analysis. At the same time, filtering algorithms are used to remove any possible random interference or abnormal fluctuations, thereby improving data stability and validity. For example, when collecting frequency and temperature data, due to the different sensor update rates, the sampling point times need to be aligned, and sudden changes need to be removed through methods such as median filtering.

[0034] Normalization and standardization are performed on the initial preprocessed sequence according to preset sampling rules. After alignment and noise reduction, the data structure is restructured according to the set data sampling period or data window length. Data of different dimensions is converted to a unified scale and distribution range, making it more suitable for subsequent modeling and analysis. Normalization is typically used to scale data to a range between 0 and 1, while standardization transforms data to a mean of 0 and a variance of 1.

[0035] The processed, normalized data is transmitted to distributed computing nodes close to the data source. Edge computing devices then perform subsequent data analysis and decision-making tasks, reducing response latency and data transmission pressure. Edge computing devices can be microprocessors or embedded intelligent control boards deployed at the resonator site, capable of performing tasks such as model calculation, parameter evaluation, and anomaly identification. Table 1 shows the most recent record of multi-dimensional resonance parameter preprocessing and distribution.

[0036] Table 1: The most recent multi-dimensional resonance parameter preprocessing and distribution record

[0037]

[0038] Furthermore, the present application also includes: extracting parameter features and resonance features of the preprocessing sequence based on the edge computing device, and inputting them into the neural network structure for training; generating a dynamic resonance behavior prediction model through the training results; performing control data prediction on the real-time preprocessing data according to the dynamic resonance behavior prediction model, and performing performance evaluation; generating compensation parameters based on the performance evaluation results and integrating them into the control unit.

[0039] Specifically, edge computing devices extract parameter and resonance characteristics from preprocessed sequences. This normalized and standardized data is then fed into an edge-side intelligent computing module to further identify representative characteristic information. Parameter characteristics primarily refer to numerical variations such as frequency change rate, amplitude fluctuation, and temperature gradient. Resonance characteristics focus on underlying properties closely related to the operating state of the quartz crystal, such as signal morphology, frequency offset trends, and resonance quality factors, reflecting actual operating conditions. Statistical analysis is combined with signal processing algorithms, such as principal component analysis or Fourier transforms, to extract the most valuable data for modeling.

[0040] The extracted feature sequence is input into a constructed deep neural network. As a multi-layered structure that mimics the neural activity of the human brain, neural networks are capable of recognizing complex nonlinear relationships and learning the mapping patterns between input features and output responses. During training, internal parameters are continuously adjusted based on the error backpropagation mechanism, gradually improving prediction accuracy. For example, by learning the linkage between amplitude changes and temperature changes, it is possible to identify the specific pattern of resonator frequency drift under temperature fluctuations.

[0041] The trained neural network structure is solidified into an executable computational logic that is used to predict the resonator's response behavior under various environmental changes in real time. This allows the dynamic relationship between the quartz crystal resonator's internal parameters and external operating conditions to be captured. Based on real-time data input, the system can output the resonant parameter change trend over the next several time steps. For example, when a rising temperature trend is detected, the quartz crystal resonator can predict the direction and degree of possible frequency shift in a few minutes.

[0042] Using a dynamic resonant behavior prediction model to predict control data and evaluate performance on real-time preprocessed data involves inputting current operating data into the model to predict optimal control response data, such as the circuit drive voltage or temperature compensation value that should be adjusted. The predicted results are then compared with actual operational feedback to assess the model's prediction accuracy and response quality. Performance evaluation is based on a combination of metrics such as error calculation and response delay determination. For example, if the average prediction error falls below a certain threshold, the prediction performance is considered good.

[0043] Generating compensation parameters based on performance evaluation results and integrating them into the control unit means determining whether the current control effect meets the target based on model performance feedback. If not, new parameter adjustment values ​​are generated. Compensation parameters such as temperature compensation voltage, bias adjustment current, or oscillation amplitude correction coefficient are written to the resonator control system in real time to further correct the system output and improve stability and accuracy. For example, if the frequency prediction error is persistently high, the temperature compensation coefficient may be increased to correct the drift.

[0044] Furthermore, the present application also includes: extracting sensitive factors from the preprocessing sequence based on the performance evaluation results, establishing a dynamically optimized sensitive mapping relationship between parameters and performance indicators; based on the real-time preprocessing data, fine-tuning the parameter characteristic data in the sensitive mapping relationship, and sending the optimized control parameters to the control unit in real time; monitoring the optimized control effect and performing feedback correction.

[0045] Specifically, extracting sensitive factors from the preprocessing sequence based on performance evaluation results means retrospectively analyzing the data sequence used for input modeling after evaluating the control model's predictive performance to identify parameters that have a strong influence on changes in system performance, called sensitive factors. Sensitive factors are characteristic variables that significantly affect resonator performance with even small changes, such as the gradient of the frequency response to temperature changes or the amplitude response to fluctuations in the driving voltage. The extraction process uses methods such as sensitivity analysis and correlation assessment to identify variables that are highly correlated with output performance indicators for subsequent optimization.

[0046] Establishing a dynamically optimized sensitive mapping relationship between parameters and performance indicators involves using identified sensitive factors as key input variables, constructing a mathematical mapping model between them and system performance indicators (such as frequency stability and quality factor changes), and ensuring that the mapping model can automatically adjust over time or as operating conditions change. This mapping relationship can be constructed using methods such as regression analysis, gradient descent, or neural networks, allowing the corresponding performance impact of any input parameter change to be accurately calculated.

[0047] Based on real-time preprocessed data, the parameter characteristic data in the sensitive mapping relationship is fine-tuned, and the optimized control parameters are sent to the control unit in real time. The newly collected multi-dimensional state data is input into the established mapping relationship model, and the current optimal control parameter combination is inferred. The sensitive factors are adjusted slightly to achieve dynamic optimization of the control behavior. The control parameters may include drive current amplitude, temperature compensation gain, bias voltage value, etc., which are immediately adopted by the control unit and acted on the quartz crystal resonator.

[0048] Monitoring the optimized control effect and performing feedback corrections means continuing to track the actual operating status after executing the optimized control instructions. By comparing performance indicators before and after optimization, such as the improvement in frequency stability or the reduction in resonance point fluctuation, the effectiveness of the optimization cycle is determined. If the expected effect is found to be below the set target, the optimization strategy will be automatically adjusted or parameters will be re-adjusted based on the feedback results to ensure that the control process is always in a positive closed loop.

[0049] Furthermore, the present application also includes: extracting the residuals of the real-time perception data and the predicted control data, inputting the residual sequence into a fault identifier to identify the fault mode corresponding to the absolute value of the residual in the residual sequence; and outputting the fault prediction result according to the fault mode.

[0050] Specifically, extracting the residual between real-time perception data and predictive control data means comparing the resonator operating status data currently collected in real time by the sensor with the control response data calculated based on the predictive model, and calculating the difference between the two. The difference is the residual, which reflects the deviation between the predicted value and the true value. It can reveal the degree of deviation from the model's prediction accuracy and the system's operation. If the residual continues to increase at a certain moment, it may indicate that the system has entered an abnormal operating state. For example, if the predicted resonant frequency should be 10 MHz, but the sensor actually measures 9.97 MHz, the residual is 0.03 MHz. Long-term accumulation may point to potential faults.

[0051] Inputting the residual sequence into a fault identifier to identify the fault pattern corresponding to the absolute value of the residual in the residual sequence involves arranging the calculated residuals into a chronological sequence and inputting it into a trained fault identifier. The identifier then determines whether the current operating state conforms to a known fault type based on characteristics such as the magnitude, trend, and duration of the residual values. A fault pattern is a characteristic template of a specific fault in data representation, such as amplitude drop, frequency drift, or short-duration pulse interference. The identification process may be based on algorithms such as neural networks, support vector machines, or decision trees, comparing and matching the residual pattern with an existing fault database to arrive at a specific judgment result.

[0052] The fault prediction results are output based on the fault pattern. When the fault identifier determines that the residual sequence matches a typical fault pattern, it outputs a predictive judgment result, indicating that the current quartz crystal resonator may be in the early stages of developing a certain fault or is about to fail. This prediction result may include information such as the fault type, probability of occurrence, and expected duration, prompting dispatchers or operations personnel to intervene in advance. The output results not only support alarm or current limiting mechanisms, but can also be used to adjust subsequent control strategies.

[0053] Furthermore, the present application also includes: introducing a residual connection structure based on the residual sequence, and designing a recursive feedback module in combination with time series information; training the response sensitivity of the sensitive factor through the attention mechanism; and using the continuous training results to adjust the parameter feature data weight distribution of the dynamic resonance behavior prediction model in the recursive feedback module.

[0054] Specifically, introducing a residual connection structure based on the residual sequence and designing a recursive feedback module in combination with time series information refers to introducing a structural design when modeling residual data that allows the model to retain a direct path between input and output during the calculation process, allowing the original information to be transmitted layer by layer within the model without being completely replaced, thereby alleviating the gradient vanishing problem in deep models and improving training stability. The role of the residual connection structure is to enhance the model's ability to remember key features, enabling the prediction model to maintain an accurate understanding of the input residual signal even at a deeper level. At the same time, combining time series information means considering the historical states of several past time steps at the same time. The recursive feedback module uses time series dependencies to continuously adjust the prediction deviation of the previous moment, enabling dynamic correction of long-term trend errors.

[0055] Training the response sensitivity of sensitive factors through the attention mechanism refers to introducing an adaptive focusing method during the model training process, which automatically identifies input variables that have a greater influence on the prediction results, thereby assigning higher computational weights to the input variables. The attention mechanism is an empowerment strategy that dynamically weights each type of input feature based on its importance in different contexts, enabling the dynamic resonant behavior prediction model to prioritize factors that have the greatest impact on the prediction results when processing complex inputs. Training the response sensitivity of sensitive factors means that the dynamic resonant behavior prediction model will continuously adjust the weight structure during the learning process, allowing it to accurately identify the features that are the basis for key decisions when faced with multi-dimensional parameter inputs.

[0056] Utilizing continuous training results to adjust the weight distribution of parameter feature data within the dynamic resonant behavior prediction model within the recursive feedback module means updating and reconstructing the model's internal parameter weights based on the error feedback and prediction results from each training round during the model's continuous training and feedback loop. The recursive feedback module resets the influence of various parameters on the prediction process, making the dynamic resonant behavior prediction model more tailored to the current operating conditions and resonator state. This updated weight distribution enables the model to continuously learn from data and maintain a high degree of sensitivity to phenomena such as abnormal patterns and trend drift.

[0057] Furthermore, the present application also includes: retrieving fault sample data; performing a similarity comparison between the fault mode and the fault sample data; generating a preliminary diagnostic conclusion based on the similarity results; calling a diagnostic mechanism to determine the fault development trend, generating a diagnostic output result including the fault type, warning level, possible location and recommended response, and adding it to the fault prediction result.

[0058] Specifically, known fault instances are extracted from a pre-built historical database or knowledge base. This sample data typically includes the characteristics of various fault types under different operating conditions, relevant sensor data change curves, environmental parameters at the time of occurrence, and the final processing results. The existence of this fault sample data provides a reference basis for comparative analysis of current fault modes, ensuring the accuracy and engineering interpretability of the identification results. For example, during the use of quartz crystal resonators, typical faults such as frequency drift, drive failure, or temperature anomalies were recorded, and the corresponding sensor output sequences and repair logs were retained in the database.

[0059] A similarity comparison is performed between the fault mode and the fault sample data. The fault mode currently identified through the residual sequence is compared and analyzed with the retrieved historical samples to determine the degree of similarity in the feature dimension. The comparison process is based on multidimensional feature matching algorithms such as Euclidean distance, dynamic time warping, or cosine similarity to quantify the similarity score between the current abnormal data and the historical samples. The higher the similarity, the more likely a known fault process is being reproduced. For example, if the current frequency drop rate is highly consistent with the temperature drift fault curve in a historical sample, it can be preliminarily determined to be of the same type.

[0060] Generating a preliminary diagnosis based on similarity results involves inferring the specific fault type currently being experienced based on highly similar samples obtained through similarity comparison, and making a first-level fault diagnosis based on this information. A preliminary diagnosis indicates a suspected fault type, but the specific severity or development trend requires further data analysis. For example, similar samples may indicate frequency attenuation corresponding to aging-related offsets, but further confirmation is needed to determine whether it is in the early or critical stages.

[0061] Invoking the diagnostic mechanism to determine the fault's development trend and generate diagnostic output including the fault type, warning level, possible location, and recommended response, involves further initiating multi-layer analysis modules. Combining current operating data with preliminary conclusions, the system assesses the fault's development speed, scope, and potential impact, thereby determining whether the fault presents a risk of continued deterioration and assigning a warning level. Warning levels can be categorized as prompt, warning, and severe. Possible location refers to the most likely location of the fault, inferred from the data, such as the circuit's temperature compensation section, resonant drive module, or packaging area. Recommended responses provide operational recommendations to maintenance personnel or the automated control system, such as parameter calibration, load switching, or shutdown inspection.

[0062] Furthermore, the present application also includes: adaptively updating the control model parameters of the dynamic resonance behavior prediction model at a learning rate according to the residual sequence; and saving a model snapshot and verifying its effectiveness during the evolution of the collaborative control strategy.

[0063] Specifically, adaptively updating the learning rate of the control model parameters of the dynamic resonant behavior prediction model based on the residual sequence means that the residual information between the measured data and the model prediction is continuously used during operation to self-adjust the key parameters in the prediction model, where the learning rate represents the step size or change rate of the parameter update. Adaptive updating means that the learning rate is not fixed, but is automatically adjusted according to the residual size, change trend or training stage, accelerating learning when the error is large and slowing down the update rate when the error stabilizes, so as to achieve more efficient and stable parameter optimization. The control model parameters include structural variables such as weights, biases, and regularization factors in the neural network, which determine the response strength and learning ability of the dynamic resonant behavior prediction model to the input. The residual-driven learning rate adjustment mechanism can improve the prediction accuracy and convergence speed under different working conditions.

[0064] During the evolution of the collaborative control strategy, model snapshots are saved and their validity verified. During the entire multi-node collaborative operation process, whenever a new control strategy is generated or a model parameter reconstruction is completed, the model state at that time will be recorded as a model snapshot. The snapshot includes the model structure, parameter configuration, current weight values, and some input and output samples, providing a benchmark for subsequent model evaluation and comparison. Verifying the validity means backtesting or cross-validating the snapshot after saving it to determine whether the dynamic resonance behavior prediction model has high control performance, stability, and generalization ability in the current or similar environment. If the verification result is good, the snapshot will be used as a candidate model for subsequent control selection; if the effect is not ideal, it will not be adopted and the control strategy will be readjusted. Collaborative control strategy evolution refers to the process in which multiple resonator nodes work together, and each model gradually forms a better global control strategy through information sharing and strategy iteration. The evolution process changes dynamically, so snapshot records have the significance of preserving key evolution paths.

[0065] Furthermore, the present application also includes: initializing the information communication link between multiple resonator nodes, and collecting the operating data of each node in real time and performing standardized processing; after standardized processing, determining the collaborative control strategy according to the operating status of each node and executing distributed control; dynamically switching the collaborative control priority according to the fault prediction results.

[0066] Specifically, in multiple quartz crystal resonators, a data transmission channel is established between the nodes to form an information communication link, which can ensure the real-time exchange of parameters, status, instructions and other information between the nodes. Initialization settings mean that at the initial startup, it is necessary to configure the identification mark, data protocol, bandwidth allocation and connection sequence of each communication chain to ensure that the communication process is stable and reliable. Subsequently, operating data such as resonant frequency, temperature, voltage, current, drive status, etc. are continuously collected from each node and standardized. Standardization refers to the conversion of data of different dimensions and dimensions into a data format with a unified scale, so that the operating information from different nodes can be compared and integrated in the same analysis model.

[0067] After completing unified data processing, the current operating status of each resonator is comprehensively analyzed, including whether it is stable, whether it is close to the boundary conditions, the current load level, and the predicted risk, and a cross-node collaborative control strategy is constructed accordingly. The collaborative control strategy is a control mechanism in which multiple nodes cooperate and complement each other. For example, when the temperature of a resonator node is high and the amplitude decreases, the control system may decide to temporarily reduce the load of the node and let another stable node take on more tasks. Distributed control means that the control instructions are not issued by a central node, but are distributed among multiple control units. After receiving local data and global strategy instructions, each node independently performs its own adjustment tasks, which improves the system's response speed and fault tolerance.

[0068] Dynamically switching collaborative control priorities based on fault prediction results means adjusting the task allocation weight or execution level of a node in the collaborative control system in real time after the fault prediction model identifies a potential anomaly or high-risk trend at a node. Control priority refers to the resource allocation authority and the priority level of control instructions when executing regulatory tasks. When a node is marked as potentially faulty, its priority decreases, automatically shifting critical or overloaded tasks to nodes in good condition. Dynamic switching means continuous evolution based on prediction results, ensuring robustness and adaptability.

[0069] Furthermore, the present application also includes: identifying the trigger conditions for decreased control accuracy, and modifying the strategy based on the trigger conditions; verifying the strategy modification results online and retaining the optimal version; constructing an evolutionary map based on historical optimization paths; and performing multi-strategy fusion based on the evolutionary map to generate an optimized control strategy.

[0070] Specifically, identifying the trigger conditions for a decrease in control accuracy and making strategy corrections based on the trigger conditions means continuously monitoring the deviation between the current control output and the target control value during continuous operation to determine whether the control accuracy meets the set requirements. When it is found that the error between the predicted value and the actual value continues to expand, or the control parameters output by the model cause a significant increase in the fluctuation of the system response, it is considered that the control accuracy has decreased. Trigger conditions refer to a set of predefined judgment criteria, such as residuals exceeding a set threshold, response delays exceeding limits, or a decrease in control stability indicators. Once any of these conditions are met, the strategy correction process will be automatically started. Strategy correction refers to updating the currently used control logic, parameter configuration, or feature weights to improve control performance.

[0071] Online verification of the strategy modification results and retention of the optimal version means that after a strategy modification is completed, it will not be immediately used as the final control solution, but will be tested in real time in the current operating environment in a parallel or alternative manner. The verification process is an online verification, and its goal is to determine whether the modified strategy is better than the current benchmark by comparing the control effects before and after the modification, such as frequency stability, response time, energy consumption fluctuations and other indicators. If the modification result is significantly better than the existing solution, the solution will be retained as the current optimal control version for continued use in subsequent operations. Retaining the optimal version means building a strategy library, so that the historical optimal solution can be quickly called when facing different working conditions.

[0072] An evolutionary map is constructed based on historical optimization paths. All historical strategy revision, verification, and retention processes are recorded and visually organized to form a dynamic structure, which, as an evolutionary map, describes the entire process of the control strategy from initial setting to multiple rounds of adjustment. It includes information such as the motivation for each revision, the content of the modification, the performance improvement effect, and the applicable conditions. The purpose of constructing an evolutionary map is to accumulate knowledge during the long-term operation of the system, identify strategy combinations that perform better in specific environments, and then assist in subsequent strategy generation and fusion decision-making. For example, the map may show that under low-temperature and low-load conditions, the frequency compensation and drive amplitude linkage strategy is superior to individual adjustments in the long term.

[0073] Based on the evolutionary graph, multiple strategies are integrated to generate an optimized control strategy. Based on the relationships between different strategy nodes in the graph, the effective components of multiple strategies are integrated to form an integrated, more adaptable, and robust control solution. The key to multi-strategy integration lies in identifying the advantages of each strategy under different conditions and combining them through weighted combination, feature sharing, and module interchange, so that the generated control strategy has higher response speed and stability under new operating conditions. For example, by integrating a temperature control strategy that was effective in high-frequency fluctuation scenarios with a bias adjustment strategy that is effective in voltage instability scenarios, the system can achieve more precise regulation and control in complex electromagnetic interference environments.

[0074] In summary, the control method of a quartz crystal resonator provided in this application has the following technical effects: achieving the technical goal of intelligent adjustment of the operating state of the quartz crystal resonator based on dynamic prediction and collaborative control, and achieving the technical effect of improving control accuracy, adaptability and overall system reliability.

[0075] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0076] Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application is intended to include these modifications and variations.

Claims

1. A method for controlling a quartz crystal resonator, characterized in that: include: Obtain a multi-dimensional resonance performance parameter sequence of a quartz crystal resonator; A dynamic resonance behavior prediction model is constructed using a multi-dimensional resonance performance parameter sequence; In combination with the dynamic resonance behavior prediction model, identifying abnormal operating status of the quartz crystal resonator and generating a fault prediction result; Fault type judgment and diagnosis output are performed based on the fault prediction results, and a collaborative control strategy is generated to execute distributed control.

2. The method for controlling a quartz crystal resonator according to claim 1, wherein: include: Performing time series alignment and noise filtering processing on the multi-dimensional resonance performance parameter sequence to obtain an initial preprocessing sequence; Normalizing and standardizing the initial preprocessing sequence according to a preset sampling rule to obtain a preprocessing sequence; Distribute the pre-processing sequence to edge computing devices.

3. The method for controlling a quartz crystal resonator according to claim 2, wherein: include: Extract parameter features and resonance features from the preprocessing sequence based on the edge computing device, and input them into a neural network structure for training; Generate a dynamic resonance behavior prediction model through training results; Performing control data prediction on the real-time preprocessed data according to the dynamic resonance behavior prediction model and performing performance evaluation; Compensation parameters are generated based on the performance evaluation results and integrated into the control unit.

4. A method for controlling a quartz crystal resonator according to claim 3, characterized in that: include: Extracting sensitive factors from the preprocessing sequence based on the performance evaluation results, and establishing a dynamically optimized sensitive mapping relationship between parameters and performance indicators; Based on the real-time pre-processed data, fine-tuning the parameter characteristic data in the sensitive mapping relationship is performed, and the optimized control parameters are sent to the control unit in real time; Monitor the optimized control effect and perform feedback correction.

5. The method for controlling a quartz crystal resonator according to claim 4, wherein: include: Extracting residuals from real-time sensing data and predictive control data, and inputting the residual sequence into a fault identifier to identify a fault mode corresponding to the absolute value of the residual in the residual sequence; A fault prediction result is output according to the fault mode.

6. A method for controlling a quartz crystal resonator according to claim 5, characterized in that: include: A residual connection structure is introduced according to the residual sequence, and a recursive feedback module is designed in combination with time series information; Training the response sensitivity of sensitive factors through attention mechanism; The continuous training results are used to adjust the parameter feature data weight distribution of the dynamic resonance behavior prediction model in the recursive feedback module.

7. A method for controlling a quartz crystal resonator according to claim 6, characterized in that: include: Retrieve fault sample data; Performing a similarity comparison between the fault mode and fault sample data; Generate preliminary diagnostic conclusions based on similar results; The diagnostic mechanism is called to determine the fault development trend, and the diagnostic output results including the fault type, warning level, possible location and recommended response are generated and added to the fault prediction results.

8. The method for controlling a quartz crystal resonator according to claim 6, wherein: include: Performing a learning rate adaptive update on the control model parameters of the dynamic resonance behavior prediction model according to the residual sequence; Model snapshots are saved and validity is verified during the evolution of the collaborative control strategy.

9. The method for controlling a quartz crystal resonator according to claim 1, wherein: include: Initialize the information communication link between multiple resonator nodes, collect and standardize the operation data of each node in real time; After standardization, the coordinated control strategy is determined according to the operating status of each node and distributed control is executed; The collaborative control priority is dynamically switched according to the fault prediction result.

10. The method for controlling a quartz crystal resonator according to claim 1, wherein: include: Identify trigger conditions for decreased control accuracy and modify the strategy based on the trigger conditions; Verify the strategy modification results online and retain the optimal version; Construct an evolutionary graph based on historical optimization paths; Multi-strategy fusion is performed according to the evolutionary graph to generate an optimized control strategy.

Images