AI-based optical module firmware verification automation integration test system

Abstract

This invention discloses an AI-based automated integrated testing system for optical module firmware verification, belonging to the field of optical communication equipment testing technology. Specific features include: deploying a three-sensor matrix of optical, electrical, and environmental sensors to synchronously collect core parameters of the optical module; setting an appropriate sampling frequency to ensure complete capture of transient fluctuations and timestamp synchronization; employing a dynamic threshold method based on temperature change rate to eliminate outliers; extracting three basic features: temporal coupling, frequency domain correlation, and spatial distribution; combining fault sensitivity coefficients and feature redundancy with weighted fusion to form comprehensive features; constructing a multi-stage fusion diagnostic system; dynamically switching and adapting sub-models; accurately identifying various faults and associating them with environmental parameters to locate the causes; the system supports incremental updates of model parameters and core parameter threshold libraries; and automatically generates standardized test reports containing feature correlation maps and parameter change curves, significantly improving the efficiency, accuracy, and intelligence level of optical module firmware verification.

Description

Technical Field

[0001] This invention belongs to the field of optical communication equipment testing technology, specifically relating to an AI-based automated integrated testing system for optical module firmware verification. Background Technology

[0002] As a core component of optical communication systems, the firmware performance of optical modules directly determines the stability, transmission rate, and reliability of the communication link. Firmware verification is a crucial step in the factory testing and lifecycle maintenance of optical modules. With the continuous increase in requirements for transmission bandwidth and response speed of optical modules in fields such as 5G and data centers, firmware functions are becoming increasingly complex. They need to cover the entire lifecycle from startup calibration, stable operation, to aging and degradation, and must cope with the coupled effects of multiple dimensions such as optical signals, electrical parameters, and environmental stress. This places stringent demands on the automation, precision, and intelligence of testing systems.

[0003] In existing technologies, firmware verification for optical modules faces several bottlenecks: First, testing is primarily manual or semi-automated, relying on operator experience to configure the testing environment and collect parameters, resulting in low efficiency and the potential for overlooking transient faults due to human error. Second, parameter acquisition is limited to a single dimension, focusing mainly on core optical parameters such as optical power and wavelength, lacking simultaneous acquisition of multi-dimensional parameters related to light, electricity, and the environment, making it difficult to capture the coupling relationships between parameters and leading to incomplete fault diagnosis. Third, outlier determination uses a fixed threshold method, failing to consider normal parameter fluctuations caused by environmental factors such as temperature and vibration, easily resulting in misjudgments or omissions. Fourth, fault diagnosis models lack lifecycle adaptability; a single model cannot adequately handle different types of faults, such as timing anomalies during startup, parameter drift during stable periods, and performance degradation during aging periods, and cannot effectively pinpoint the cause of the fault. Fifth, the testing system lacks versatility and adaptability, making it difficult to adapt to firmware version iterations of different optical module models; adding new fault cases requires re-tuning the model, resulting in high maintenance costs.

[0004] Therefore, there is an urgent need for an AI-based automated integrated testing system for optical module firmware verification to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-based automated integrated testing system for optical module firmware verification, which solves the technical problems in the prior art where fault diagnosis models lack lifecycle adaptability and do not consider normal parameter fluctuations caused by environmental factors.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] An AI-based automated integrated testing system for optical module firmware verification includes:

[0008] The multi-dimensional parameter synchronous acquisition module deploys a three-sensor matrix of optical, electrical, and environmental sensors to synchronously acquire core parameters such as wavelength drift and transient fluctuations of optical power throughout the entire life cycle of the optical module, sets an appropriate minimum effective sampling frequency, and completes the timestamp synchronization of the acquired data.

[0009] The outlier identification module uses a dynamic threshold method based on the temperature change rate to identify and remove outliers. It extracts three basic features: time-domain coupling, frequency-domain correlation, and spatial distribution. The fusion weight is determined by calculating the fault sensitivity coefficient and feature redundancy, and finally, the comprehensive features are obtained.

[0010] The multi-stage AI fusion diagnostic module constructs a multi-stage model fusion diagnostic system, adapts and dynamically switches corresponding sub-models according to the startup phase, stable phase, and aging phase, outputs fault types, generates a visual traceability report, and incrementally updates the model and parameter threshold library.

[0011] Furthermore, a three-sensor matrix of optical, electronic, and environmental sensors is deployed, specifically using the following method:

[0012] The optical-electrical-environmental triple sensor matrix includes an optical signal acquisition sub-matrix, an electrical parameter acquisition sub-matrix, and an environmental stress acquisition sub-matrix. The three work together to achieve synchronous acquisition of multi-dimensional parameters.

[0013] The optical signal acquisition sub-matrix includes a high-precision spectrometer or wavelength meter and a high-speed response optical power meter. The high-precision spectrometer or wavelength meter is used to monitor the center wavelength of the optical signal output by the optical module in real time. By comparing it with the standard rated wavelength, the dynamic drift difference is calculated to obtain the wavelength drift λ. The high-speed response optical power meter is used to acquire the instantaneous value of optical power in real time. By calculating the difference between the instantaneous values ​​of optical power at adjacent acquisition times, the transient fluctuation Pt of optical power is obtained.

[0014] The electrical parameter acquisition sub-matrix includes a high-frequency current probe, an oscilloscope, and a differential probe. The high-frequency current probe is connected in series in the optical module bias circuit and is used with the oscilloscope to acquire the current waveform. The fluctuation amplitude value of the high-frequency ripple component is extracted to obtain the bias current ripple Ir. The differential probe is used with a high-speed oscilloscope to acquire the SerDes high-speed serial signal and generate an eye diagram. The peak jitter is analyzed through the eye diagram to obtain the SerDes signal jitter Jp.

[0015] The environmental stress acquisition sub-matrix includes a temperature sensor, a triaxial accelerometer, an EMI receiver, and a near-field probe. The temperature sensor is mounted on a key location of the optical module and calculates the real-time temperature change rate Vt by the time difference between adjacent sampling points. The triaxial accelerometer is attached to the outer shell of the optical module and collects vibration acceleration data ax, ay, and az in the XYZ directions. The EMI receiver works in conjunction with the near-field probe to measure the electromagnetic radiation intensity Em at fixed test points around the optical module.

[0016] Furthermore, a minimum effective sampling frequency is set to synchronize the timestamps of the collected data. The specific method is as follows:

[0017] The minimum effective sampling frequency is determined based on the maximum rate of change of the core parameters of the optical module. The sampling frequency of the data collected synchronously from the core parameters is not lower than the minimum effective sampling frequency, and the sampling frequency range needs to ensure complete capture of transient parameter fluctuations within a preset unit. Timestamp synchronization is achieved through system clock calibration. All data collected by the sensors carry timestamps of uniform precision to ensure the temporal correlation of multi-dimensional parameters.

[0018] Furthermore, temporal coupling features are extracted, specifically using the following method:

[0019] Calculate the slope of the coordinated change of any two parameters ,in This represents the change in the duration parameter x of time period T1, that is, the absolute value of the difference between the parameter value at the start time and the parameter value at the end time of the time period. This represents the change in the parameter y for time duration T1. This represents the mean of the duration parameter x for time T1. The mean value of the time parameter y is represented by the slope of the coordinated change among different types of parameters, which characterizes the linkage effect between parameters.

[0020] Furthermore, frequency domain correlation features are extracted, specifically using the following method:

[0021] A joint spectral analysis of the bias current ripple Ir and the electromagnetic radiation intensity Em was performed, using the formula... Extract the cross-power spectral density, where T represents the duration of the spectral analysis. This represents the mathematical expectation operator, used to calculate the statistical average of the expression within parentheses. This represents the frequency domain expression of the bias current ripple Ir after Fast Fourier Transform. For frequency, The conjugate complex number representing the frequency domain expression of electromagnetic radiation intensity Em. This represents the cross-power spectral density values ​​of the bias current ripple and electromagnetic radiation intensity at different frequencies f.

[0022] Furthermore, spatial distribution features are extracted, specifically using the following method:

[0023] Based on triaxial vibration acceleration Using the sensor position coordinates, the formula Calculate the spatial distribution entropy of vibration energy and quantify the non-uniform effect of vibration on the internal circuitry of the module. This represents the probability density of vibration acceleration in direction i, where i corresponds to the X, Y, and Z measurement directions of the triaxial accelerometer.

[0024] Furthermore, the fusion weights are determined by calculating the fault sensitivity coefficient and feature redundancy, ultimately yielding the comprehensive features. The specific method is as follows:

[0025] Define a set of basic features and extract n independent basic features, denoted as F1, F2, ..., Fn;

[0026] For the random variable of the u-th basic feature Fu and the corresponding optical module firmware fault tag Through mutual information formula Calculate the fault sensitivity coefficient , The larger the value, the stronger the ability of the basic characteristic Fu to distinguish faults. ,in, It is a random variable. Let represent the set of fault labels, with values ​​representing discrete categories. y represents the specific value of the fault label, which is the actual value taken by the random variable Ya. f represents the specific value of the basic feature Fu. This represents the set of all possible values ​​for the basic feature Fu. This represents the probability that the random variable Ya takes the value y. This represents the probability that the basic feature Fu takes the value f. This indicates the fault label when the basic feature Fu takes the value f. The joint probability of y;

[0027] Calculate the u-th basic feature Fu with all other basic features. The mean cosine similarity is used to obtain the feature redundancy Ru. The larger Ru is, the more information Fu has that is repeated with other features.

[0028] Define the fusion weight Wu of the u-th basic feature Fu, and use the formula It means that among them Represents the v-th basic feature The corresponding fault sensitivity coefficient, Represents the v-th basic feature The corresponding feature redundancy is then fused to obtain the comprehensive feature. .

[0029] Furthermore, a multi-stage model fusion diagnostic system is constructed, specifically through the following methods:

[0030] Using comprehensive features as input, a multi-stage model fusion diagnostic system is constructed. The model architecture design is as follows: For the full lifecycle characteristics of optical modules, three stages of sub-model adaptation are implemented.

[0031] Startup phase: A lightweight CNN model is adopted, with an input layer → convolutional layer 1 → pooling layer (2×2 max pooling) → convolutional layer 2 → fully connected layer (128 neurons) → output layer structure. The components in the comprehensive input feature F that are strongly correlated with fast convergence characteristics are identified, and the focus is on identifying wavelength calibration algorithm failure, abnormal timing configuration and other startup phase-specific faults.

[0032] Stable period: Enable the XGBoost model, using the complete integrated features F as input, and employ cross-validation during training;

[0033] Aging period: Using the GRU time series model, a comprehensive feature sequence of continuous duration is input to predict the probability of failure in the future duration and identify parameter decay failures caused by firmware aging in advance.

[0034] Furthermore, the fault type is output using the following method:

[0035] Edge computing nodes receive comprehensive features F in real time and call the corresponding sub-models according to the current operating stage; after the sub-models output the probability distributions of various faults, they are used to determine the faults using the fault determination formula. Calculate the final failure type, where s represents the s-th stage. For stage weights, Let F be the probability that the comprehensive feature F belongs to fault type c under stage s.

[0036] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0037] 1. This invention utilizes a three-sensor matrix of optical, electrical, and environmental sensors to synchronously acquire core parameters throughout the entire process of optical module startup, stabilization, and aging. It covers multiple dimensions including optical signals, electrical parameters, and environmental stress. Combined with a sampling frequency design adapted to the rate of parameter change, it comprehensively captures millisecond-level transient fluctuations and high-frequency anomalies. The system requires no manual intervention throughout, automatically completing parameter acquisition, timestamp synchronization, and data preprocessing. This completely eliminates the efficiency bottlenecks and oversight risks of traditional manual testing, significantly improving test throughput and meeting the needs of large-scale optical module verification.

[0038] 2. This invention employs a dynamic threshold method based on temperature change rate to effectively distinguish between normal parameter fluctuations caused by the environment and actual fault jumps, avoiding misjudgments and missed diagnoses caused by fixed thresholds. By extracting three types of features—temporal coupling, frequency domain correlation, and spatial distribution—and combining them with a weighted fusion of fault sensitivity and feature redundancy, a multi-stage diagnostic system adapted to the entire lifecycle is constructed to accurately identify different types of faults, such as timing anomalies during startup, parameter drift during the stable period, and performance degradation during the aging period. Simultaneously, it generates a visual source tracing report by associating environmental parameters, intuitively presenting parameter changes and feature correlations before and after the fault, facilitating rapid location of the root cause of the fault.

[0039] 3. This invention supports an incremental learning mechanism, automatically integrating new test data to update model parameters monthly, and optimizing the core parameter threshold library quarterly based on qualified module test data. This continuously adapts to optical module firmware version iterations and model expansions, ensuring compatibility with new scenarios without repetitive debugging. Standardized test reports automatically integrate test data, fault conditions, and model performance indicators, enabling automated delivery of test results and reducing system maintenance costs and the barrier to entry. Whether for small- to medium-batch R&D verification or large-scale mass production testing, it maintains highly efficient and stable testing performance, significantly improving the intelligence and scalability of optical module firmware verification. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 The diagram shows the modular design of the AI-based automated integrated testing system for optical module firmware verification according to the present invention.

[0042] Figure 2 The diagram illustrates the steps of the AI-based automated integrated testing method for optical module firmware verification according to the present invention.

[0043] Figure 3 The diagram shows the AI-based optical module firmware quick connection structure of the present invention;

[0044] Figure 4 The diagram shows the AI-based optical module firmware quick connection structure of the present invention;

[0045] Reference numerals: Cage assembly 1, Module assembly 2, Socket assembly 3, Baffle assembly 4, Jumper assembly 5. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1, such as Figure 1 , Figure 2 The AI-based automated integration testing system for optical module firmware verification shown here includes the following steps:

[0048] The multi-dimensional parameter synchronous acquisition module deploys a three-sensor matrix of optical, electrical, and environmental sensors to synchronously acquire core parameters such as wavelength drift and transient fluctuations of optical power throughout the entire life cycle of the optical module, sets an appropriate minimum effective sampling frequency, and completes the timestamp synchronization of the acquired data.

[0049] Deploy a three-sensor matrix of optical, electronic, and environmental sensors to achieve synchronous acquisition of core parameters throughout the entire lifecycle of the optical module, from startup and stable operation to aging and degradation. The core parameters acquired include:

[0050] Using a high-precision spectrometer or wavelength meter, the center wavelength of the optical signal output from the optical module is monitored in real time and compared with the standard rated wavelength. The dynamic drift difference is calculated to obtain the wavelength drift, which is denoted as [missing value]. It is used to characterize the dynamic change of optical signal wavelength from the standard value. Based on historical data and industry standards, the wavelength drift threshold is set in combination with the center wavelength of the optical module. s (e.g., 0.5nm);

[0051] A high-speed response optical power meter is used to collect the instantaneous value of optical power in real time. The transient fluctuation of optical power is obtained by calculating the difference between the instantaneous values ​​of optical power at adjacent collection times, which is denoted as Pt. This is used to capture the instantaneous fluctuation characteristics of optical power in milliseconds or even microseconds.

[0052] A high-frequency current probe was connected in series in the bias circuit, and the current waveform was acquired with an oscilloscope. The fluctuation amplitude value of the high-frequency ripple component was extracted to obtain the bias current ripple Ir. A stable test environment at room temperature was set up (temperature 25℃±2℃, humidity 45%-65%, no electromagnetic interference). 100 qualified optical modules of the same model were selected, and each module was run continuously for 24 hours. The bias current waveform was acquired with a high-frequency current probe (bandwidth ≥1GHz) and an oscilloscope (sampling rate ≥10GS / s). The maximum ripple value of each module was extracted, and the 95th percentile of the 100 maximum values ​​was calculated (5% extreme values ​​were removed to avoid the influence of individual differences). Finally, the standard threshold of bias current ripple was determined.

[0053] A high-speed oscilloscope with a differential probe is used to acquire the SerDes high-speed serial signal, generate an eye diagram and analyze the peak jitter to obtain the SerDes signal jitter Jp, which is used to describe the timing offset of the high-speed serial signal.

[0054] Attach PT100 or thermocouple sensors to key locations on the module to collect temperature data in real time. Calculate the rate of change by the time difference between adjacent sampling points to determine the real-time temperature change rate Vt.

[0055] A triaxial accelerometer is attached to the housing of the optical module to collect vibration acceleration data in the X, Y, and Z directions. This reflects the vibration intensity experienced by the module;

[0056] Using an EMI receiver with a near-field probe, fixed test points are set around the module to measure the electromagnetic radiation intensity Em, which is used to quantify the electromagnetic interference level.

[0057] The minimum effective sampling frequency is determined based on the maximum rate of change of the core parameters of the optical module. The sampling frequency of the data collected synchronously with the core parameters is set to be greater than or equal to the minimum effective sampling frequency to ensure that transient parameter fluctuations can be fully captured. The collected data is then timestamped.

[0058] The outlier identification module uses a dynamic threshold method based on the temperature change rate to identify and remove outliers. It extracts three basic features: time-domain coupling, frequency-domain correlation, and spatial distribution. The fusion weight is determined by calculating the fault sensitivity coefficient and feature redundancy, and finally, the comprehensive features are obtained.

[0059] To address the dynamic identification and removal of outliers, an outlier detection method based on dynamic adjustment of temperature change rate is introduced. A dynamic threshold for outlier detection is set, which is the average of any parameter value x and its T1 duration (T1 duration is set according to actual needs). When the absolute value of the deviation exceeds the dynamic threshold Dy for anomaly detection, i.e. If a value is found to be abnormal, its timestamp and associated environmental parameters are recorded simultaneously. This is used to distinguish between normal parameter fluctuations caused by environmental stress and genuine abnormal jumps caused by equipment failure or interference. The formula is then used to... This represents the dynamic threshold for anomaly detection, where The standard deviation of the corresponding parameter x represents its normal fluctuation range. The real-time temperature change rate is represented by c, which is a constant coefficient obtained based on historical data from the stable phase of qualified optical modules of the same model. In this embodiment, the 3σ distribution of normal parameter fluctuations is used as the basis (covering 99.7% of normal data). Combined with reverse verification of historical abnormal cases, iterative fine-tuning is performed to finally determine the optimal benchmark value (typically 2.8-3.2) for this type of optical module. k represents the temperature change sensitivity. With the goal of correctly eliminating fluctuations caused by normal temperature and not missing sudden abnormalities, the reasonable range of k values ​​is traversed to select the k value that achieves the best judgment balance effect. Represents the natural constant.

[0060] Determine the temporal coupling characteristics and calculate the slope of the coordinated change of any two parameters (such as the transient fluctuation of optical power Pt and the rate of temperature change Vt). ,in This represents the change in the duration parameter x of time period T1, that is, the absolute value of the difference between the parameter value at the start time and the parameter value at the end time of the time period. This represents the change in the parameter y for time duration T1. This represents the mean of the duration parameter x for time T1. The mean value of the T1 duration parameter y is represented by the slope of the coordinated change among different types of parameters, which characterizes the linkage effect between parameters.

[0061] To determine the frequency domain correlation characteristics, a joint spectral analysis of the bias current ripple Ir and the electromagnetic radiation intensity Em was performed, using the formula... Extract the cross-power spectral density, where T represents the duration of the spectral analysis. This represents the mathematical expectation operator, used to calculate the statistical average of the expression within parentheses. This represents the frequency domain expression of the bias current ripple Ir after Fast Fourier Transform (FFT). For frequency, The conjugate complex number representing the frequency domain expression of electromagnetic radiation intensity Em (used to characterize the frequency domain correlation between the two). This represents the cross-power spectral density values ​​of the bias current ripple and electromagnetic radiation intensity at different frequencies f.

[0062] Determine spatial distribution characteristics based on triaxial vibration acceleration Using the sensor position coordinates, the formula Calculate the spatial distribution entropy of vibration energy and quantify the non-uniform effect of vibration on the internal circuitry of the module. This represents the probability density of vibration acceleration in direction i, where i corresponds to the X, Y, and Z measurement directions of the triaxial accelerometer.

[0063] Define a set of basic features and extract n independent basic features (denoted as F1, F2, ..., Fn), covering three core dimensions, such as the cooperative slope of the transient fluctuation of optical power and the rate of temperature change denoted as F1, the cross-power spectral density of bias current ripple and electromagnetic radiation intensity denoted as F2, and the energy distribution entropy of triaxial vibration acceleration denoted as F3.

[0064] For the random variable of the u-th basic feature Fu and the corresponding optical module firmware fault tag Through mutual information formula Calculate the fault sensitivity coefficient , The larger the value, the stronger the ability of the basic characteristic Fu to distinguish faults. ,in, It is a random variable. Let represent the set of fault labels, with values ​​representing discrete categories, such as {normal, wavelength calibration failure, SerDes timing anomaly, power management failure, ...}. y represents the specific value of the fault label, which is the actual value of the random variable Ya. f represents the specific value of the basic feature Fu. This represents the set of all possible values ​​for the basic feature Fu. If it is a continuous feature, it is discretized into a finite number of intervals. This represents the probability that the random variable Ya takes the value y. This represents the probability that the basic feature Fu takes the value f. This indicates the fault label when the basic feature Fu takes the value f. The joint probability of y;

[0065] Calculate the u-th basic feature Fu with all other basic features. The mean cosine similarity is used to obtain the feature redundancy Ru. The larger Ru is, the more information Fu has that is repeated with other features.

[0066] Define the fusion weight Wu of the u-th basic feature Fu, and use the formula It means that among them Represents the v-th basic feature The corresponding fault sensitivity coefficient, Represents the v-th basic feature The corresponding feature redundancy is then fused to obtain the comprehensive feature. ;

[0067] The multi-stage AI fusion diagnostic module constructs a multi-stage model fusion diagnostic system, adapts and dynamically switches corresponding sub-models according to the startup phase, stable phase, and aging phase, outputs fault types, generates a visual traceability report, and incrementally updates the model and parameter threshold library.

[0068] Using comprehensive features as input, a multi-stage model fusion diagnostic system is constructed;

[0069] Model architecture design: Based on the characteristics of the entire lifecycle of optical modules, a three-stage adaptation sub-model is implemented:

[0070] Startup Phase: A lightweight CNN model is used, employing the following structure: Input layer (feature dimensions adapted to the core components of the comprehensive features, such as 8-dimensional) → Convolutional layer 1 (64 3×3 convolutional kernels, ReLU activation) → Pooling layer (2×2 max pooling) → Convolutional layer 2 (32 3×3 convolutional kernels, ReLU activation) → Fully connected layer (128 neurons) → Output layer (number of fault categories). The Adam optimizer is used with a learning rate of 0.001, training for 30 epochs, and an early stopping strategy (stopping if the validation set accuracy does not improve for 5 consecutive epochs). Inference latency is compressed to ≤10ms using TensorRT quantization. Components in the comprehensive input feature F that are strongly correlated with fast convergence characteristics (such as temporal coupling features and SerDes jitter-related features) are selected. Inference latency is controlled to ≤10ms. The focus is on identifying startup-specific faults such as wavelength calibration algorithm failure and abnormal timing configuration.

[0071] Stable period: The XGBoost model is enabled, with the complete integrated feature F as input. In addition to the preset hyperparameters (learning rate 0.1, tree depth 6), the minimum number of sample leaf nodes is 3 and the regularization coefficient λ=0.1 to prevent overfitting. Cross-validation is used during training.

[0072] Aging period: A GRU time series model is adopted, with a sequence length of 60 (corresponding to 1 hour of continuous comprehensive feature data), a hidden layer dimension of 64, and a dropout rate of 0.2. The AdamW optimizer is used with a learning rate of 0.0005 and a training epoch of 50. The continuous comprehensive feature sequence is input to predict the probability of failure in the future time period and identify parameter decay failures caused by firmware aging in advance.

[0073] The system sets up phase identification logic, such as real-time monitoring of the optical module's runtime and core parameter stability. During the startup phase (0-3 minutes), runtime is used as the trigger condition. When the temperature change rate Vt ≤ 0.1℃ / min and the transient fluctuation of optical power Pt ≤ 0.01dBm within 5 consecutive minutes, it automatically switches to the stable phase (3 minutes-23 hours). When the runtime is ≥ 23 hours or the parameter decay trend (such as wavelength drift λ continuously approaching the threshold λs) is ≥ 5% within 1 consecutive hour, it switches to the aging phase.

[0074] Edge computing nodes receive comprehensive features F in real time and call the corresponding sub-models according to the current operating stage; after the sub-models output the probability distributions of various faults, they are used to determine the faults using the fault determination formula. Calculate the final failure type, where s represents the s-th stage. The weights are assigned to different stages (0.3 for the startup phase, 0.5 for the stable phase, and 0.2 for the aging phase). Let F be the probability that the comprehensive feature F belongs to fault type c under stage s.

[0075] Based on the fault tracing data package, a visual report is automatically generated, including the core parameter change curves 10 minutes before and after the fault (such as wavelength drift λ, bias current ripple Ir, etc.), time-domain / frequency-domain feature correlation spectrum (such as the cooperative slope change of Pt and Vt, the cross power spectral density of Ir and Em), and the spatial distribution entropy heat map of vibration energy; by associating with the environmental parameters recorded in step two (temperature, humidity, electromagnetic interference Em), the fault cause is located (such as excessive environmental electromagnetic interference causing abnormal SerDes timing).

[0076] New test data (including unidentified fault cases) is automatically summarized every month, and after being labeled by testers, it is added to the training set to retrain the three sub-models. The model parameters are updated through incremental learning. The constant coefficient c and temperature change sensitivity k in the dynamic threshold Dy for anomaly identification are optimized simultaneously to ensure that the threshold is adapted to firmware version iterations and changes in the test environment.

[0077] Each quarter, based on the 24-hour stable operation data of 100 newly added qualified optical modules, the threshold values ​​of core parameters such as bias current ripple and wavelength drift are recalculated. After removing extreme values, the 95th percentile is taken to update the system threshold library and improve the accuracy of parameter determination.

[0078] After the test is completed (either the full life cycle test of a single optical module is completed or the test is terminated according to the preset test duration), the system automatically integrates the test data, fault conditions, and model performance indicators (accuracy, recall, inference latency) of each stage, generates a standardized test report, and realizes automated delivery of test results.

[0079] Example 2, as follows Figure 3 , Figure 4 The AI-based optical module firmware rapid connection structure shown consists of five core parts. These components work together to simultaneously meet the requirements for electrical connection stability, mechanical structural strength, and electromagnetic interference resistance. Specifically, they include:

[0080] Core load-bearing components: cage assembly 1; functional execution components: module assembly 2; electrical connection components: socket assembly 3; auxiliary function components: baffle assembly 4 and jumper assembly 5 (optional).

[0081] Basic assembly sequence:

[0082] Step 1: Solder and fix the socket assembly 3 onto the motherboard, ensuring that the positioning pins and the positioning holes on the motherboard are fully aligned;

[0083] Step 2: Fix the cage assembly 1 to the equipment housing by installing the mounting ears, and adjust the position so that the cage cavity is coaxially aligned with the socket assembly 3;

[0084] Step 3: Insert module component 2 into the cage cavity from right to left along the guide groove until the gold fingers (a key technical feature of electronic connection, usually used for circuit board connection) are fully inserted into the socket component 3;

[0085] Step 4: Pass the screws through the washer, module screw hole, and cage positioning hole (if there is a baffle, the baffle clips must be fixed to the opening of the cage before passing the screws through), and tighten clockwise to complete the fixing.

[0086] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0087] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An AI-based automated integrated testing system for optical module firmware verification, characterized in that, include: The multi-dimensional parameter synchronous acquisition module deploys a three-sensor matrix of optical, electrical, and environmental sensors to synchronously acquire core parameters such as wavelength drift and transient fluctuations of optical power throughout the entire life cycle of the optical module, sets an appropriate minimum effective sampling frequency, and completes the timestamp synchronization of the acquired data. The outlier identification module uses a dynamic threshold method based on the temperature change rate to identify and remove outliers. It extracts three basic features: time-domain coupling, frequency-domain correlation, and spatial distribution. The fusion weight is determined by calculating the fault sensitivity coefficient and feature redundancy, and finally, the comprehensive features are obtained. Define a set of basic features and extract n independent basic features, denoted as F1, F2, ..., Fn; For the random variable of the u-th basic feature Fu and the corresponding optical module firmware fault tag Through mutual information formula Calculate the fault sensitivity coefficient , The larger the value, the stronger the ability of the basic characteristic Fu to distinguish faults. ,in, It is a random variable. Let represent the set of fault labels, with values ​​representing discrete categories. y represents the specific value of the fault label, which is the actual value taken by the random variable Ya. f represents the specific value of the basic feature Fu. This represents the set of all possible values ​​for the basic feature Fu. This represents the probability that the random variable Ya takes the value y. This represents the probability that the basic feature Fu takes the value f. This indicates the fault label when the basic feature Fu takes the value f. The joint probability of y; Calculate the u-th basic feature Fu with all other basic features. The mean cosine similarity is used to obtain the feature redundancy Ru. The larger Ru is, the more information Fu has that is repeated with other features. Define the fusion weight Wu of the u-th basic feature Fu, and use the formula It means that among them Represents the v-th basic feature The corresponding fault sensitivity coefficient, Represents the v-th basic feature The corresponding feature redundancy is then fused to obtain the comprehensive feature. ; The multi-stage AI fusion diagnostic module constructs a multi-stage model fusion diagnostic system, adapts and dynamically switches corresponding sub-models according to the startup phase, stable phase, and aging phase, outputs fault types, generates a visual traceability report, and incrementally updates the model and parameter threshold library. Using comprehensive features as input, a multi-stage model fusion diagnostic system is constructed. The model architecture design is as follows: For the full lifecycle characteristics of optical modules, three stages of sub-model adaptation are implemented. Startup phase: A lightweight CNN model is used, and the components in the comprehensive feature F that are strongly correlated with fast convergence characteristics are input to identify startup-specific faults such as wavelength calibration algorithm failure and abnormal timing configuration. Stable period: Enable the XGBoost model, using the complete integrated features F as input, and employ cross-validation during training; Aging period: Using the GRU time series model, inputting a comprehensive feature sequence of continuous duration, predicting the probability of failure within the future duration, and identifying parameter decay failures caused by firmware aging in advance; Edge computing nodes receive comprehensive features F in real time and call the corresponding sub-model according to the current operating stage. After the sub-model outputs the probability distribution of various faults, the final fault type is calculated by the fault determination formula, where s represents the s-th stage, is the stage weight, and is the probability that comprehensive feature F belongs to fault type c under stage s.

2. The AI-based automated integrated testing system for optical module firmware verification according to claim 1, characterized in that, The specific method for deploying a combined optical, electronic, and environmental sensor matrix is ​​as follows: The optical-electrical-environmental triple sensor matrix includes an optical signal acquisition sub-matrix, an electrical parameter acquisition sub-matrix, and an environmental stress acquisition sub-matrix. The three work together to achieve synchronous acquisition of multi-dimensional parameters. The optical signal acquisition sub-matrix includes a high-precision spectrometer or wavelength meter and a high-speed response optical power meter. The high-precision spectrometer or wavelength meter is used to monitor the center wavelength of the optical signal output by the optical module in real time. By comparing it with the standard rated wavelength, the dynamic drift difference is calculated to obtain the wavelength drift λ. The high-speed response optical power meter is used to acquire the instantaneous value of optical power in real time. By calculating the difference between the instantaneous values ​​of optical power at adjacent acquisition times, the transient fluctuation Pt of optical power is obtained. The electrical parameter acquisition sub-matrix includes a high-frequency current probe, an oscilloscope, and a differential probe. The high-frequency current probe is connected in series in the optical module bias circuit and is used with the oscilloscope to acquire the current waveform. The fluctuation amplitude value of the high-frequency ripple component is extracted to obtain the bias current ripple Ir. The differential probe is used with a high-speed oscilloscope to acquire the SerDes high-speed serial signal and generate an eye diagram. The peak jitter is analyzed through the eye diagram to obtain the SerDes signal jitter Jp. The environmental stress acquisition sub-matrix includes a temperature sensor, a triaxial accelerometer, an EMI receiver, and a near-field probe. The temperature sensor is mounted on a key location of the optical module and calculates the real-time temperature change rate Vt by the time difference between adjacent sampling points. The triaxial accelerometer is attached to the outer shell of the optical module and collects vibration acceleration data ax, ay, and az in the XYZ directions. The EMI receiver works in conjunction with the near-field probe to measure the electromagnetic radiation intensity Em at fixed test points around the optical module.

3. The AI-based automated integrated testing system for optical module firmware verification according to claim 1, characterized in that, Set the minimum effective sampling frequency for adaptation and complete the timestamp synchronization of the collected data. The specific method is as follows: The minimum effective sampling frequency is determined based on the maximum rate of change of the core parameters of the optical module. The sampling frequency of the data collected synchronously from the core parameters is not lower than the minimum effective sampling frequency, and the sampling frequency range needs to ensure complete capture of transient parameter fluctuations within a preset unit. Timestamp synchronization is achieved through system clock calibration. All data collected by the sensors carry timestamps of uniform precision to ensure the temporal correlation of multi-dimensional parameters.

4. The AI-based automated integrated testing system for optical module firmware verification according to claim 1, characterized in that, The specific method for extracting temporal coupling features is as follows: Calculate the slope of the coordinated change of any two parameters ,in This represents the change in the duration parameter x of time period T1, that is, the absolute value of the difference between the parameter value at the start time and the parameter value at the end time of the time period. This represents the change in the parameter y for time duration T1. This represents the mean of the duration parameter x for time T1. The mean value of the time parameter y is represented by the slope of the coordinated change among different types of parameters, which characterizes the linkage effect between parameters.

5. The AI-based automated integrated testing system for optical module firmware verification according to claim 1, characterized in that, Extracting frequency domain correlation features, the specific method is as follows: A joint spectral analysis of the bias current ripple Ir and the electromagnetic radiation intensity Em was performed, using the formula... Extract the cross-power spectral density, where T represents the duration of the spectral analysis. This represents the mathematical expectation operator, used to calculate the statistical average of the expression within parentheses. This represents the frequency domain expression of the bias current ripple Ir after Fast Fourier Transform. For frequency, The conjugate complex number representing the frequency domain expression of electromagnetic radiation intensity Em. This represents the cross-power spectral density values ​​of the bias current ripple and electromagnetic radiation intensity at different frequencies f.

6. The AI-based automated integrated testing system for optical module firmware verification according to claim 1, characterized in that, The specific method for extracting spatial distribution features is as follows: Based on triaxial vibration acceleration Using the sensor position coordinates, the formula Calculate the spatial distribution entropy of vibration energy and quantify the non-uniform effect of vibration on the internal circuitry of the module. This represents the probability density of vibration acceleration in direction i, where i corresponds to the X, Y, and Z measurement directions of the triaxial accelerometer.

Images