Debugging device and method for AI human sensor, and AI human sensor

Abstract

The application provides an AI human body sensor debugging device and method and an AI human body sensor. The application belongs to the technical field of intelligent sensing. The method comprises the following steps: constructing a special debugging device and building a comprehensive debugging environment; placing the AI human body sensor in the comprehensive debugging environment, performing preliminary hardware parameter setting, obtaining initial hardware parameter configuration data, starting a physiological signal reproduction module and an interference environment simulation module, simultaneously collecting original output data of the sensor in the comprehensive debugging environment, performing preliminary analysis on the original output data, and generating a preliminary performance deviation analysis report. The automatic batch debugging process design reduces tedious operations such as manual repeated plugging and unplugging and visual waveform judgment, improves the debugging efficiency, and guarantees the consistency of sensor debugging on a large-scale production line.

Description

Technical Field

[0001] This invention proposes an AI human body sensor debugging device, method, and AI human body sensor, belonging to the field of intelligent sensing technology. Background Technology

[0002] In the field of AI human body sensors, such as PPG photoelectric sensors, MEMS accelerometers, thermopile infrared temperature sensors, and bioimpedance electrodes, the debugging process before mass production and deployment is crucial. However, current debugging technologies have many prominent problems.

[0003] Traditional calibration is mostly conducted in static laboratories, relying solely on standard signal sources or simple analog circuits, which cannot reproduce the nonlinearity, time-varying nature, and individual variability of real human physiological signals. Key factors such as the influence of skin color on PPG signals and the attenuation of bioimpedance due to subcutaneous fat thickness are not considered, resulting in a significant decrease in performance of factory-specified parameters during actual wear.

[0004] Meanwhile, the lack of a controllable, quantifiable, and composite injection mechanism for typical interference sources prevents the sensor's anti-interference capability from being fully verified. Furthermore, the independent tuning of parameters for each hardware module, without considering system-level coupling effects, results in suboptimal overall performance.

[0005] Furthermore, the debugging process relies on repeated manual plugging and unplugging and visual waveform judgment, which is not only time-consuming and inconsistent but also difficult to meet the needs of large-scale production lines. Existing public information and manufacturer documentation often describe vague concepts and fail to start from the physical layer of sensor hardware to design dedicated debugging devices and standardized processes. In essence, it uses software to cover up hardware defects. Therefore, there is an urgent need for a new debugging method that focuses on sensor hardware performance. Summary of the Invention

[0006] This invention provides an AI human body sensor debugging device, method, and AI human body sensor to solve the problems mentioned in the background art above:

[0007] This invention proposes a debugging method for an AI human body sensor, the method comprising:

[0008] S1. Construct a dedicated debugging device and set up a comprehensive debugging environment;

[0009] S2. Place the AI ​​human body sensor in the comprehensive debugging environment, perform preliminary hardware parameter settings, and obtain initial hardware parameter configuration data; and start the physiological signal reproduction module and interference environment simulation module, while collecting the sensor's raw output data in the comprehensive debugging environment; perform preliminary analysis on the raw output data and generate a preliminary performance deviation analysis report;

[0010] S3. Based on the preliminary performance deviation analysis report, enter the hardware parameter collaborative debugging stage; generate a hardware parameter collaborative adjustment scheme through intelligent algorithms, collaboratively adjust the sensor hardware parameters, and collect sensor output data again to generate collaboratively adjusted output data;

[0011] S4. Perform in-depth analysis on the output data after collaborative adjustment and conduct performance evaluation to generate a performance improvement evaluation report. If the performance improvement does not meet the preset target, further optimize the hardware parameter collaborative adjustment scheme according to the performance improvement evaluation report, and repeat S3 and S4 until the sensor performance reaches the preset standard and generate the final hardware parameter optimization data.

[0012] S5. Based on the final hardware parameter optimization data, design an automated batch debugging process for AI human body sensors; generate an automated batch debugging process plan, perform batch debugging of AI human body sensors on large-scale production lines, and generate batch debugging completion data; conduct random checks and verifications on the batch debugging completion data, and generate a final debugging qualification report.

[0013] The present invention proposes an AI human body sensor debugging device, which includes a memory, a processor, and a computer program stored in the memory and executable on the memory. The processor executes the program to implement the AI ​​human body sensor debugging method as described above.

[0014] The present invention proposes an AI human body sensor, which stores a computer program that is executed by a processor to implement the debugging method of the AI ​​human body sensor as described above.

[0015] The beneficial effects of this invention are as follows: By constructing a dedicated debugging device to reproduce real human physiological signals and typical interference environments, the physiological authenticity and interference controllability of sensor debugging are greatly improved, allowing the sensor to adapt to complex and ever-changing real-world application scenarios before leaving the factory. During the debugging process, a three-stage collaborative debugging process is adopted, fully considering system-level coupling effects and precisely optimizing key sensor hardware parameters. This significantly enhances the sensor's sensitivity, recognition accuracy, and anti-interference capability, effectively reducing the suboptimal overall performance issues caused by isolated parameter tuning. Simultaneously, the automated batch debugging process design reduces tedious manual operations such as repeated plugging and unplugging and visual waveform judgment, not only improving debugging efficiency but also ensuring consistency in sensor debugging on large-scale production lines. Furthermore, this method strictly focuses on sensor hardware debugging, avoiding the problem of using software to mask hardware defects. It achieves collaborative optimization of core sensor indicators from the hardware physical layer and ensures stable and reliable performance of the sensor in practical applications, providing users with a better user experience. Attached Figure Description

[0016] Figure 1This is a diagram illustrating the steps of the method described in this invention. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0018] One embodiment of the present invention, such as Figure 1 As shown, a method for debugging an AI human body sensor includes:

[0019] S1. Construct a dedicated debugging device, which includes a physiological signal reproduction module and an interference environment simulation module. The physiological signal reproduction module reproduces the nonlinear, time-varying, and individual-differentiated characteristics of real human physiological signals through multi-parameter adjustable physiological excitation sources, generating simulated real human physiological signal data. The interference environment simulation module can controllably, quantitatively, and compositely inject typical interference sources, including 50Hz power frequency electromagnetic interference, motion artifacts, sudden changes in ambient light, and temperature and humidity fluctuations, generating simulated interference environment data. Based on the simulated real human physiological signal data and the simulated interference environment data, a comprehensive debugging environment is built.

[0020] S2. Place the AI ​​human body sensor in the comprehensive debugging environment, perform preliminary hardware parameter settings, and obtain initial hardware parameter configuration data; based on the initial hardware parameter configuration data, start the physiological signal reproduction module and the interference environment simulation module, and at the same time collect the sensor's raw output data in the comprehensive debugging environment; perform preliminary analysis on the raw output data, identify performance deviation problems caused by debugging environment distortion or lack of interference simulation, and generate a preliminary performance deviation analysis report.

[0021] S3. Based on the preliminary performance deviation analysis report, the hardware parameter collaborative debugging stage begins. In this stage, the parameters of each hardware module (preamplifier, filter, ADC) are no longer tuned in isolation. Instead, the system-level coupling effect is considered, with sensitivity, accuracy, and anti-interference as the three core indicators for optimization. A hardware parameter collaborative adjustment scheme is generated through intelligent algorithms. According to the hardware parameter collaborative adjustment scheme, the sensor hardware parameters are collaboratively adjusted, and the sensor output data is collected again to generate the collaboratively adjusted output data.

[0022] S4. Perform in-depth analysis on the output data after collaborative adjustment to evaluate the performance improvement of the sensor in terms of sensitivity, recognition accuracy and anti-interference capability, and generate a performance improvement evaluation report. If the performance improvement does not meet the preset target, further optimize the hardware parameter collaborative adjustment scheme according to the performance improvement evaluation report, and repeat steps S3 and S4 until the sensor performance reaches the preset standard, and generate the final hardware parameter optimization data.

[0023] S5. Based on the final hardware parameter optimization data, design an automated batch debugging process for AI human body sensors. This process uses a dedicated debugging device and automated control software to automatically load sensor hardware parameters, automatically switch debugging environments, automatically evaluate performance and provide feedback adjustments, generating an automated batch debugging process plan. Following the automated batch debugging process plan, perform batch debugging of AI human body sensors on a large-scale production line, generating batch debugging completion data. Randomly check and verify the batch debugging completion data to ensure that each batch of sensors meets the preset performance standards, generating a final debugging qualification report.

[0024] The working principle and effects of the above technical solution are as follows:

[0025] This debugging method constructs a comprehensive debugging environment that closely resembles real-world scenarios, accurately reproducing physiological signal characteristics and complex interferences, significantly improving debugging accuracy and avoiding performance misjudgments caused by environmental simulation distortion. It employs a hardware parameter collaborative debugging mode, breaking the limitations of isolated optimization, fully considering the coupling effect between modules, enhancing parameter adaptability, reducing the time cost of repeated trial and error, and significantly improving core performance aspects such as sensor sensitivity, accuracy, and anti-interference capability. Through an iterative optimization mechanism, performance is continuously verified, ensuring that each sensor meets preset standards, effectively improving product qualification rates. The automated batch debugging process, combined with dedicated devices and software, significantly reduces manual debugging costs and human error, significantly improving production line debugging efficiency. The sampling inspection and verification stage after batch debugging accurately intercepts unqualified products, preventing defective products from entering the market and enhancing product reliability.

[0026] In one embodiment of the present invention, S1 includes:

[0027] S11. Based on the application scenarios of AI human body sensors (such as medical monitoring and sports fitness), determine the types of physiological signals that need to be reproduced (such as heart rate, blood oxygen, electromyography, electroencephalography, etc.) and interference scenarios, and obtain real physiological signal data of different age groups and physical conditions from existing information databases to build a physiological signal feature database containing 100,000+ samples. At the same time, organize the quantitative parameter range of typical interferences (such as 50Hz power frequency interference intensity 0-10V / m).

[0028] S12. Build a dedicated debugging device that integrates a physiological signal reproduction module, an interference environment simulation module, a signal fusion module and a data acquisition module. The physiological signal reproduction module is equipped with a 0-5V adjustable excitation source and a signal waveform synthesis unit. The interference environment simulation module is equipped with an electromagnetic interference generator, a motion artifact simulation robotic arm, an ambient light controller and a temperature and humidity control box.

[0029] S13. Perform zero-point calibration and gain calibration on the physiological signal reproduction module. Based on the physiological signal feature database, generate nonlinear time-varying signals containing diurnal rhythm changes and motion state switching through the waveform synthesis unit, and superimpose individual difference parameters (such as the heart rate fluctuation range of the elderly) to generate a multi-scenario simulated real human physiological signal dataset.

[0030] S14. Perform precision calibration on each interference unit (e.g., ambient light change amplitude error ≤ 5%), and support independent output of a single interference or composite output of multiple interferences according to a preset ratio (e.g., power frequency interference + motion artifact composite scene), generating a simulated interference environment dataset covering 20+ typical interference scenarios.

[0031] S15. Use the signal fusion module to output physiological signals and interference signals synchronously at the millisecond level and build a comprehensive debugging environment; select 3 standard calibration sensors to connect to the environment, collect output data and compare it with theoretical values ​​to ensure that the signal distortion is ≤3%, and lock the environmental parameters to verify the effectiveness of the environment.

[0032] The working principle and effects of the above technical solution are as follows:

[0033] By constructing a physiological signal feature database of over 100,000 samples, covering real data from different age groups and physical conditions, and clearly defining the quantitative parameter range of typical interferences, the realism of physiological signal and interference simulations is significantly improved, avoiding the problem of one-sided simulation scenarios caused by single samples or ambiguous interference parameters. The dedicated debugging device integrates multiple modules, eliminating the hassle of disparate device configurations, enhancing the integration and convenience of debugging, and reducing the risk of device compatibility failures. Precise calibration of physiological signals and interference units ensures controllable simulation signal distortion; for example, the amplitude error of ambient light abrupt changes is ≤5%, significantly improving the accuracy of simulation data and avoiding inaccurate debugging benchmarks due to module deviations. Supporting single or compound interference outputs, generating data from 20+ typical scenarios, and combined with millisecond-level signal synchronization, the comprehensive debugging environment is closer to actual application scenarios, enhancing the targeting of debugging. Verifying the effectiveness of the environment using standard sensors ensures signal distortion is ≤3%, preventing misjudgments in subsequent debugging due to unreliable environments.

[0034] In one embodiment of the present invention, S13 includes:

[0035] Based on the specifications of the physiological signal reproduction module, a zero-point and gain calibration process was developed. A standard signal source was used for calibration, and a calibration confirmation report was generated upon completion.

[0036] Based on the calibration qualification confirmation report, the module is started, and the physiological signal feature database is called to extract the basic amplitude and frequency parameters of relevant signals (such as heart rate and blood oxygen) and generate a basic physiological signal waveform set.

[0037] The basic physiological signal waveform set is input into the waveform synthesis unit, loaded with the diurnal rhythm time sequence curve and the motion state switching threshold, and superimposed to generate a nonlinear time-varying signal dataset.

[0038] The individual difference parameter library (including age and physical condition correlation coefficient) is retrieved from the physiological signal feature database, and time-varying signals are matched one by one and parameters are superimposed to generate signal data with individual characteristics.

[0039] Signal data with individual characteristics are classified and labeled according to scenarios. After removing abnormal data, they are integrated to generate a dataset of real human physiological signals simulating multiple scenarios.

[0040] The working principle and effects of the above technical solution are as follows:

[0041] By calibrating the module according to a standard signal source and generating a qualified report, the output accuracy of the physiological signal reproduction module is ensured from the source. This avoids the distortion of the basic signal caused by module zero drift and gain deviation, laying a solid foundation for subsequent signal generation. Core parameters such as heart rate and blood oxygen are extracted from the database to generate basic waveforms, ensuring the signals closely match real physiological characteristics. This reduces the randomness of generating signals out of thin air and improves the authenticity of the basic signals. Adding and superimposing circadian rhythms and motion state thresholds gives the signals realistic nonlinear time-varying characteristics of the human body, breaking the limitations of fixed waveforms. Matching individual parameters such as age and physical condition imbues the signals with unique characteristics, enhancing adaptability to different populations and avoiding the problem of a single signal being unable to simulate diverse physiological states. Finally, classifying, labeling, and removing abnormal data further improves the reliability of the dataset, ensuring that the output multi-scenario simulated signals accurately match actual debugging needs.

[0042] In one embodiment of the present invention, S2 includes:

[0043] S21. Perform visual inspection and power-on self-test on the AI ​​human body sensor to be debugged, and remove any residual interference sources on the sensor surface; fix the sensor in the standard test station of the integrated debugging environment, and calibrate the distance between the sensor and the signal transmitter using a laser positioning device (error ≤ 2mm).

[0044] S22. Load the sensor's factory default parameters into the host computer to complete the basic parameter settings. The basic parameters include sampling frequency (500Hz-1kHz), preamplifier gain (100-1000 times), and low-pass filter cutoff frequency (50-100Hz). Then, match the basic optimization parameters from the historical debugging process according to the sensor model to generate the initial hardware parameter configuration dataset.

[0045] S23. Based on the initial hardware parameter configuration data, start the three typical scenarios of the integrated debugging environment in sequence. The three typical scenarios include static physiological signal + no interference, dynamic physiological signal + single interference and dynamic physiological signal + compound interference. Collect raw output data for 20 minutes continuously in each scenario, record environmental parameters synchronously, and generate initial output datasets for multiple scenarios.

[0046] S24. Denoise (remove outliers, smooth out) and extract time-domain (peak, period) and frequency-domain (power spectral density) features from the initial output dataset of multiple scenarios, compare it one by one with the standard templates in the physiological signal feature database, and calculate the core indicators of the signal, including similarity and signal-to-noise ratio.

[0047] S25. Based on the index comparison results, identify the type of deviation (such as signal-to-noise ratio <30dB under power frequency interference, peak recognition error >5% under dynamic signals), locate the root cause (such as unreasonable filter cutoff frequency, insufficient amplification gain), and generate a preliminary performance deviation analysis report containing deviation details and root cause analysis after classification and summary.

[0048] The working principle and effects of the above technical solution are as follows:

[0049] By visually inspecting and performing a power-on self-test to eliminate residual interference, and using laser positioning to control the distance error between the sensor and the transmitter within 2mm, interference from installation deviations and initial abnormalities was eliminated at the source, improving the stability of subsequent data acquisition. After loading factory default parameters, historical optimized parameters were matched to generate the initial configuration, which better reflects the actual characteristics of the sensor than using only factory parameters, reducing invalid debugging caused by unreasonable initial parameters and improving the adaptability of parameter configuration. Three typical scenarios were sequentially started to collect 20 minutes of data and record environmental parameters, covering core scenarios from static interference-free to dynamic composite interference, avoiding the omission of performance issues due to a single test scenario and enhancing the representativeness of the initial output data. After denoising the data, time-domain and frequency-domain features were extracted and compared with standard templates to accurately calculate similarity, signal-to-noise ratio, and other indicators, reducing subjective errors in manual analysis and improving the accuracy of data interpretation. The final deviation analysis report clearly identified the type and root cause of the deviation; for example, the root cause of insufficient signal-to-noise ratio under power frequency interference was unreasonable filtering parameters, avoiding blind trial and error in subsequent debugging and providing clear guidance for parameter optimization.

[0050] In one embodiment of the present invention, S3 includes:

[0051] S31. Analyze the preliminary performance deviation analysis report, break down the core deviation indicators (sensitivity, accuracy, and immunity) into the related parameters of each hardware module (such as the preamplifier gain affecting sensitivity and the filter parameters affecting immunity); combine with the sensor hardware specifications to define the adjustable range of each parameter (such as ADC sampling bit depth 12-16 bits) and establish a parameter and indicator correlation model.

[0052] S32. Select the improved particle swarm optimization algorithm (introducing an adaptive adjustment mechanism for inertial weights), use the parameter and index correlation model as the algorithm constraint, set sensitivity ≥95%, accuracy error ≤2%, and anti-interference signal-to-noise ratio ≥40dB as the optimization objectives, and construct a hardware parameter collaborative optimization model.

[0053] S33. Input the initial deviation data into the optimization model, and generate 10 sets of hardware parameter coordinated adjustment schemes through 100 iterations; use the analytic hierarchy process to evaluate the schemes (weights: anti-interference 40%, accuracy 35%, sensitivity 25%), and select the 3 candidate schemes with the highest comprehensive scores.

[0054] S34. Select three sensors of the same model and adjust their parameters using three candidate schemes respectively. Collect and output data under a complex interference scenario. Compare the performance indicators of the three sets of data and determine the hardware parameter collaborative adjustment scheme with the best overall performance.

[0055] S35. By debugging the host computer, the parameters of the optimal solution are synchronously loaded into the sensor to be debugged, the integrated debugging environment is restarted and the output data of the whole scene is collected to generate the output dataset after collaborative adjustment.

[0056] The working principle and effects of the above technical solution are as follows:

[0057] The deviation analysis report breaks down core indicators into hardware parameters and establishes relationships, clarifying the adjustable range of parameters in conjunction with the specifications. This effectively avoids the problem of isolated parameter tuning ignoring module coupling, making parameter optimization more targeted. An improved particle swarm optimization algorithm with adaptive inertia weights is adopted, using clear targets and parameter constraints such as sensitivity ≥95% to build the model. This makes the optimization process more precise, reduces invalid iterations, and significantly improves optimization efficiency. After generating 10 schemes, the analytic hierarchy process (AHP) is used to select 3 candidate schemes based on factors such as anti-interference capability and accuracy, avoiding the one-sidedness of a single scheme and enhancing the reliability of the schemes. Through comparative testing of 3 identical sensors in a complex interference scenario, the optimal scheme is not only theoretically feasible but also closely matches actual usage needs, improving the practicality of the scheme. Finally, the optimal parameters are loaded and full-scene data is collected to ensure that the adjusted sensor meets the standards in various scenarios, avoiding the problem of partial performance being acceptable while overall performance is insufficient.

[0058] In one embodiment of the present invention, S33 includes:

[0059] Retrieve preliminary performance deviation data and parameter-index correlation model, determine the input boundary of the optimization model, configure the initial value and adjustment threshold of the inertia weight for 100 iterations, and generate the model input configuration table;

[0060] Import the input configuration table into the hardware parameter co-optimization model, start iterative calculation, output intermediate parameter combinations every 20 iterations, and finally generate 10 complete hardware parameter co-adjustment schemes.

[0061] Based on the priority of sensor applications, the evaluation dimensions of the analytic hierarchy process (AHP) were determined, and the weight allocation rules of 40% for immunity, 35% for accuracy, and 25% for sensitivity were clarified to build a solution evaluation framework.

[0062] Substitute the 10 adjustment schemes into the evaluation framework, calculate the individual scores and comprehensive weighted scores of each group in the dimensions of anti-interference, accuracy, and sensitivity, and generate a scheme score ranking table;

[0063] Based on the score ranking table, the top 3 solutions were selected, and the parameters of each solution were checked to see if they met the hardware specification limits. Finally, 3 candidate adjustment solutions were determined.

[0064] The working principle and effects of the above technical solution are as follows:

[0065] By retrieving deviation data and correlating it with the model to determine the input boundaries, and configuring iterative parameters to generate an input configuration table, the model input chaos or blind parameter settings are avoided, providing a clear basis for optimization calculations and significantly improving input accuracy. After importing the configuration table into the model, 100 iterations are performed, with intermediate results output every 20 iterations, ultimately generating 10 sets of solutions. This allows for real-time monitoring of the iteration process to promptly identify anomalies, while also ensuring solution diversity and avoiding the limitations of a single solution. The weights of the analytic hierarchy process (AHP) are determined by combining application priorities, and an evaluation framework is built with clear rules such as 40% for robustness and 35% for accuracy. This ensures that the solution evaluation closely aligns with actual needs, preventing weight imbalances that could cause the selected solution to deviate from the core requirements, thus enhancing the relevance of the evaluation. The 10 sets of solutions are quantified and ranked within the framework, using data to support comparisons, reducing subjective errors from manual selection and improving the fairness and accuracy of solution selection. After selecting the top 3, the parameters are checked to ensure they comply with the hardware specifications, preventing selected solutions from being unfeasible due to out-of-range parameters and ensuring the feasibility of the candidate solutions.

[0066] In one embodiment of the present invention, S34 includes:

[0067] Three AI human body sensors of the same model with consistent performance (factory error ≤3%) were selected, and their appearance was cleaned and self-tested upon power-on. After removing abnormal devices, they were marked as test samples 1-3, and a test sample confirmation list was generated.

[0068] According to the parameter configuration requirements of the three candidate schemes, the corresponding parameters were loaded for the three test samples in sequence through the debugging host computer. After loading for each sample, the parameter writing success rate was verified (≥99%), and a parameter configuration confirmation table was generated.

[0069] Start the composite interference scenario (preset power frequency interference + motion artifact + temperature and humidity fluctuation combination) in the comprehensive debugging environment, control the stable output of scenario parameters, and simultaneously collect 30 minutes of continuous output data from 3 test samples to generate grouped test datasets.

[0070] Extract the core indicators of sensitivity, accuracy, and anti-interference signal-to-noise ratio from the group test dataset, calculate the comprehensive performance score of each scheme according to the preset weights, and generate a multi-scheme performance comparison analysis table;

[0071] Based on the comparative analysis table and the actual application scenario requirements, the solution that meets the anti-interference standard is selected first. If multiple standards are met, the one with the highest comprehensive score is selected, and the optimal hardware parameter coordination adjustment solution is finally determined.

[0072] The working principle and effects of the above technical solution are as follows:

[0073] Sensors of the same model with a factory error ≤3% were selected, cleaned, self-tested, marked, and a confirmation list was generated. This effectively eliminated the interference of performance differences between samples on the test results, laying a fair foundation for scheme comparison and improving the reliability of subsequent tests. After loading parameters according to the candidate schemes, the success rate of writing was verified to be ≥99%, and a confirmation table was generated. This avoided test failures caused by parameter writing errors and reduced the amount of unnecessary work from repeated debugging. Composite interference scenarios such as power frequency interference and motion artifacts were started, and data was collected synchronously for 30 minutes. This complex scenario, which is close to actual use, can more accurately test the comprehensive adaptability of the schemes than single interference tests. The generated grouped datasets make the performance comparison of the schemes more intuitive. Core indicators were extracted and comprehensive scores were calculated according to preset weights. Quantitative data replaced subjective judgment, and the generated comparative analysis table clearly presented the advantages and disadvantages of each scheme, greatly reducing human error in scheme selection. Based on actual application needs, priority was given to ensuring that the anti-interference performance met the standards, and then the scheme with the highest comprehensive score was selected. This avoided selecting the theoretically optimal scheme but with poor practical adaptability, ensuring that the final optimal scheme truly fits the sensor's usage scenario.

[0074] In one embodiment of the present invention, step S4 includes:

[0075] S41. Based on the application scenario requirements of the sensor, refine the performance evaluation standards (e.g., medical-grade sensor accuracy error ≤1%, motion scene anti-interference signal-to-noise ratio ≥45dB); establish the allowable evaluation error range (e.g., ≤2%) based on the output data of the standard calibrated sensor.

[0076] S42. Perform multi-dimensional analysis on the output dataset after collaborative adjustment, quantify sensitivity using signal response time (≤100ms is acceptable), calculate physiological feature recognition accuracy using confusion matrix, evaluate anti-interference capability using interference suppression ratio (≥45dB is acceptable), verify reliability using long-term stability test (output fluctuation ≤1% for 2 consecutive hours), and generate a performance improvement evaluation report.

[0077] S43. Compare the evaluation report with the detailed standards. If there are any indicators that do not meet the standards (such as accuracy of 89% < 95% under motion artifacts + power frequency interference), test the impact of each hardware parameter on the indicator one by one by using the controlled variable method to locate the root cause (such as the filter parameter not matching the motion artifact frequency).

[0078] S44. Based on the root cause localization results, correct the target weights of the particle swarm optimization algorithm (increase the motion artifact suppression weights to 50%), and regenerate the hardware parameter coordination adjustment scheme; fine-tune key parameters (such as the filter cutoff frequency) to generate the corrected parameter set.

[0079] S45, repeat the parameter loading and data acquisition process of S35, and the analysis and evaluation process of this step, until all performance indicators meet the refined standards, lock in the final hardware parameters, and generate the final hardware parameter optimization data.

[0080] The working principle and effects of the above technical solution are as follows:

[0081] Based on detailed evaluation standards for specific scenarios, such as medical-grade accuracy error ≤1% and motion scene anti-interference signal-to-noise ratio ≥45dB, and using standard calibrated sensor data as a benchmark, performance evaluation is made more aligned with actual usage needs, improving the relevance of the evaluation. Multi-dimensional analysis is performed on the adjusted data, using response time to quantify sensitivity, confusion matrix to calculate accuracy, and interference suppression ratio to evaluate anti-interference performance. A 2-hour stability test is also added. The generated evaluation report comprehensively presents sensor performance, avoiding performance misjudgments caused by single-index evaluation and enhancing the completeness of the evaluation. When standards are not met, the control variable method is used to test the influence of each parameter one by one, accurately locating the root cause, such as mismatched filter parameters and motion artifact frequencies, avoiding ineffective trial and error by blindly adjusting parameters and significantly reducing optimization time. Based on the root cause correction algorithm weights, a new scheme is generated, and key parameters are fine-tuned, making optimization more focused on the core problem and improving the accuracy of parameter optimization. The parameter loading, acquisition, and evaluation process is repeated until standards are met, locking in the final parameters to ensure that all sensor performance indicators meet scenario standards, avoiding usage failures caused by delivery based on "partial compliance."

[0082] In one embodiment of the present invention, S44 includes:

[0083] Extract key influencing factors from the root cause localization results (such as the mismatch between motion artifact frequency and filter parameters), determine the core direction of algorithm correction, and generate a list of algorithm adjustment requirements.

[0084] Based on the requirements list, the target weights of the improved particle swarm optimization algorithm are adjusted (e.g., the motion artifact suppression weight is increased to 50%), the iterative constraints are reset, and the algorithm is run to generate a new hardware parameter coordination adjustment scheme.

[0085] Select key parameters (such as filter cutoff frequency and amplification gain) from the new adjustment scheme, determine the parameter fine-tuning range (such as filter cutoff frequency ±5Hz) based on the root cause analysis results, and generate parameter fine-tuning scheme.

[0086] The key parameters were aligned and adjusted one by one according to the fine-tuning plan. The parameter values ​​before and after the adjustment were recorded simultaneously. The parameters were verified to meet the requirements of the hardware specifications, and a draft of the revised parameter set was generated.

[0087] The initial draft of the parameter set was simulated and tested to verify its effectiveness in improving unmet indicators (such as accuracy). Once confirmed to be effective, it was locked as the revised parameter set.

[0088] The working principle and effects of the above technical solution are as follows:

[0089] Extracting key factors for root cause localization and generating a list of algorithm adjustment requirements allows for precise identification of the core issues that fail to meet standards. For example, identifying a mismatch between motion artifact frequency and filtering parameters prevents the algorithm from going astray during correction, providing a clear target for optimization and improving the targeting of adjustments. Adjusting the algorithm's target weights based on these requirements, such as increasing the motion artifact suppression weight to 50%, and then resetting constraints to generate new solutions allows optimization to focus on the weakest indicators, reducing ineffective iterations on irrelevant parameters and significantly improving the efficiency and accuracy of solution generation. Selecting key parameters such as the filtering cutoff frequency from the new solutions and defining fine-tuning ranges like ±5Hz based on root cause analysis avoids the tediousness and risks of large-scale full parameter adjustments, precisely targeting the problem points and enhancing the controllability of parameter adjustments. After adjustment, verifying whether the parameters conform to the hardware specifications, generating a draft, and then conducting simulation tests to verify the improvement effect on the non-compliant indicators ensures that the parameters are both effective and feasible, avoiding repeated adjustments due to parameters being out of range or insufficient in effect, and reducing trial-and-error costs.

[0090] In one embodiment of the present invention, step S5 includes:

[0091] S51. Based on the final hardware parameter optimization data, identify the core steps of batch debugging (parameter configuration, scene debugging, performance testing, deviation correction); use Python to write automated debugging scripts, integrate sensor batch communication interfaces (RS485 bus), environmental scene switching commands, and performance indicator calculation functions to generate an automated batch debugging process solution.

[0092] S52. Select 10 sensors for small-batch trial debugging to verify the script parameter loading success rate (≥99%) and scene switching stability; expand the debugging scene library based on common production line problems (add extreme scenarios such as low temperature environment and high humidity environment) and optimize the script fault tolerance mechanism.

[0093] S53. Deploy the optimized automated batch debugging process to the production line debugging system. Through the linkage interface between the dedicated debugging device and the production line, realize the automated connection of sensor loading, debugging and unloading. Perform batch parameter configuration and multi-scenario debugging for large-scale production line sensors, record the debugging data of each sensor in real time, and generate a batch debugging completion dataset.

[0094] S54. A stratified sampling inspection model of initial inspection + re-inspection is adopted. The initial inspection extracts 5% of the samples from the batch data and tests the core indicators in the production line self-inspection environment. The re-inspection extracts 30% of the samples that passed the initial inspection and conducts full-scenario repeated testing in an independent third-party debugging environment to verify performance consistency.

[0095] S55. Summarize the batch debugging data and sampling inspection results to generate a final debugging qualification report containing batch information, performance statistics, and sampling inspection report; and enter the debugging data of each sensor into the blockchain system.

[0096] The working principle and effects of the above technical solution are as follows:

[0097] Based on optimal parameters, core processes are analyzed and automated scripts are written. Integrated RS485 batch communication and scene switching functions generate process solutions, replacing repetitive manual operations with automation. This significantly improves batch debugging efficiency while reducing errors in manual parameter configuration, preventing debugging failures due to manual mistakes. Small-batch trial debugging verifies script success rate (≥99%), optimizes fault tolerance mechanisms, and expands the library of extreme scenarios such as low temperatures, preventing failures caused by script vulnerabilities or missing scenarios during large-scale deployment, thus enhancing process stability. After deployment to the production line, automated connection between loading, debugging, and unloading is achieved, batch parameter configuration and multi-scenario debugging are completed, and data is recorded in real time, reducing the cost of manual intervention on the production line and improving adaptability for large-scale production. The tiered sampling inspection mode of initial inspection + re-inspection is very practical. Initial inspection quickly screens, while re-inspection uses a third-party environment for full-scenario verification to ensure consistent performance, avoiding missed defects by a single sampling method and significantly improving product pass rate. After summarizing data and generating a pass report, the data of each sensor is entered into the blockchain for traceability. Parameters, performance, and other information are traceable throughout the entire lifecycle, preventing subsequent failures from being untraceable and enhancing product quality credibility.

[0098] According to one embodiment of the present invention, an AI human body sensor debugging device is characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the memory, wherein the processor executes the program to implement the AI ​​human body sensor debugging method as described above.

[0099] According to one embodiment of the present invention, an AI human body sensor has a computer program stored thereon, characterized in that the program is executed by a processor to implement the debugging method of the AI ​​human body sensor as described above.

[0100] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for debugging an AI human body sensor, characterized in that, The method includes: S1. Construct a dedicated debugging device and set up a comprehensive debugging environment; S2. Place the AI ​​human body sensor in the comprehensive debugging environment, perform preliminary hardware parameter settings, and obtain initial hardware parameter configuration data; and start the physiological signal reproduction module and interference environment simulation module, while collecting the sensor's raw output data in the comprehensive debugging environment; perform preliminary analysis on the raw output data and generate a preliminary performance deviation analysis report; S3. Based on the preliminary performance deviation analysis report, enter the hardware parameter collaborative debugging stage; generate a hardware parameter collaborative adjustment scheme through an improved particle swarm optimization algorithm, collaboratively adjust the sensor hardware parameters, and collect sensor output data again to generate collaboratively adjusted output data; S4. Perform in-depth analysis of the output data after collaborative adjustment and conduct performance evaluation to generate a performance improvement evaluation report; if the performance improvement does not meet the preset target, further optimize the hardware parameter collaborative adjustment scheme based on the performance improvement evaluation report; including: S41. Based on the requirements of sensor application scenarios, refine the performance evaluation standards; establish the allowable range of evaluation error using the output data of standard calibrated sensors as a benchmark. S42. Perform multi-dimensional analysis on the output dataset after collaborative adjustment, quantify sensitivity using signal response time, calculate physiological feature recognition accuracy using confusion matrix, evaluate anti-interference capability using interference suppression ratio, verify reliability using long-term stability test, and generate a performance improvement evaluation report. S43. Compare the evaluation report with the detailed standards. If there are any indicators that do not meet the standards, use the controlled variable method to test the impact of each hardware parameter on the indicator one by one to locate the root cause. S5. Repeat S3 and S4 until the sensor performance reaches the preset standard and generate the final hardware parameter optimization data; based on the final hardware parameter optimization data, design an automated batch debugging process for the AI ​​human body sensor; generate an automated batch debugging process plan, perform batch debugging of the AI ​​human body sensor on a large-scale production line, and generate batch debugging completion data; conduct random checks and verifications on the batch debugging completion data and generate the final debugging qualification report. S4 further includes S44, which includes: Extract the key influencing factors from the root cause localization results, determine the core direction of algorithm correction, and generate a list of algorithm adjustment requirements; Based on the requirements list, the target weights of the improved particle swarm optimization algorithm are adjusted, the iterative constraints are reset, and the algorithm is run to generate a new hardware parameter coordination adjustment scheme. From the newly adjusted plan, key parameters are selected, and the parameter fine-tuning range is determined in combination with the root cause analysis results to generate a parameter fine-tuning plan. The key parameters were aligned and adjusted one by one according to the fine-tuning plan. The parameter values ​​before and after the adjustment were recorded simultaneously. The parameters were verified to meet the requirements of the hardware specifications, and a draft of the revised parameter set was generated. The initial draft of the parameter set was simulated and tested to verify its improvement effect on the unmet indicators. After confirming its effectiveness, it was locked as the revised parameter set.

2. The debugging method for the AI ​​human body sensor according to claim 1, characterized in that, S1 includes: S11. Based on the application scenarios of AI human body sensors, determine the types of physiological signals that need to be reproduced and the interference scenarios, and obtain real physiological signal data of different age groups and physical conditions from the existing information database to construct a physiological signal feature database, while sorting out the quantitative parameter range of typical interferences. S12. Build a dedicated debugging device that integrates a physiological signal reproduction module, an interference environment simulation module, a signal fusion module, and a data acquisition module; S13. Perform zero-point calibration and gain calibration on the physiological signal reproduction module. Based on the physiological signal feature database, generate nonlinear time-varying signals containing diurnal rhythm changes and motion state switching through the waveform synthesis unit. Superimpose individual difference parameters to generate a multi-scenario simulated real human physiological signal dataset. S14. Perform precision calibration on each interference unit, and support independent output of a single interference or composite output of multiple interferences according to a preset ratio to generate a simulated interference environment dataset. S15. The physiological signal and interference signal are synchronously output at the millisecond level through the signal fusion module to build a comprehensive debugging environment; three standard calibration sensors are selected to be connected to the environment, the output data is collected and compared with the theoretical value, and the environmental parameters are locked to verify the effectiveness of the environment.

3. The debugging method for the AI ​​human body sensor according to claim 2, characterized in that, S13 includes: Based on the specifications of the physiological signal reproduction module, a zero-point and gain calibration process was developed. A standard signal source was used for calibration, and a calibration confirmation report was generated upon completion. Based on the calibration qualification confirmation report, the module is started, and the physiological signal feature database is called to extract the basic amplitude and frequency parameters of relevant signals and generate a basic physiological signal waveform set. The basic physiological signal waveform set is input into the waveform synthesis unit, loaded with the diurnal rhythm time sequence curve and the motion state switching threshold, and superimposed to generate a nonlinear time-varying signal dataset. The individual difference parameter library is retrieved from the physiological signal feature database, and time-varying signals are matched one by one and parameters are superimposed to generate signal data with individual characteristics. Signal data with individual characteristics are classified and labeled according to scenarios. After removing abnormal data, they are integrated to generate a dataset of real human physiological signals simulating multiple scenarios.

4. The debugging method for the AI ​​human body sensor according to claim 1, characterized in that, S2 includes: S21. Perform visual inspection and power-on self-test on the AI ​​human body sensor to be debugged, and remove any residual interference sources on the sensor surface; fix the sensor in the standard test station of the integrated debugging environment, and calibrate the distance between the sensor and the signal transmitter using a laser positioning device; S22. Load the sensor's factory default parameters into the host computer to complete the basic parameter settings, and then match the basic optimization parameters from the historical debugging process according to the sensor model to generate the initial hardware parameter configuration dataset. S23. Based on the initial hardware parameter configuration data, start the three typical scenarios of the integrated debugging environment in sequence, continuously collect raw output data for 20 minutes for each scenario, record environmental parameters synchronously, and generate initial output datasets for multiple scenarios. S24. Denoise the initial output dataset of multiple scenarios, extract time-domain and frequency-domain features, compare it one by one with the standard template in the physiological signal feature database, and calculate the core indicators of the signal. S25. Based on the index comparison results, identify the types of deviations, locate the root causes, and generate a preliminary performance deviation analysis report after classification and summarization.

5. The debugging method for the AI ​​human body sensor according to claim 1, characterized in that, The S3 includes: S31. Analyze the preliminary performance deviation analysis report, break down the core deviation indicators into the associated parameters of each hardware module; combine the sensor hardware specifications to define the adjustable range of each parameter and establish a parameter and indicator correlation model. S32. Using the improved particle swarm optimization algorithm, the parameter and index correlation model is used as the algorithm constraint. The optimization objectives are set as sensitivity ≥95%, accuracy error ≤2%, and anti-interference signal-to-noise ratio ≥40dB. A hardware parameter collaborative optimization model is constructed. S33. Input the initial deviation data into the optimization model, and generate 10 sets of hardware parameter coordinated adjustment schemes through 100 iterations; use the analytic hierarchy process to evaluate the schemes and select the 3 candidate schemes with the highest comprehensive scores. S34. Select three sensors of the same model and adjust their parameters using three candidate schemes respectively. Collect and output data under a complex interference scenario. Compare the performance indicators of the three sets of data and determine the hardware parameter collaborative adjustment scheme with the best overall performance. S35. By debugging the host computer, the parameters of the optimal solution are synchronously loaded into the sensor to be debugged, the integrated debugging environment is restarted and the output data of the whole scene is collected to generate the output dataset after collaborative adjustment.

6. The debugging method for the AI ​​human body sensor according to claim 5, characterized in that, S33 includes: Retrieve preliminary performance deviation data and parameter-index correlation model, determine the input boundary of the optimization model, configure the initial value and adjustment threshold of the inertia weight for 100 iterations, and generate the model input configuration table; Import the input configuration table into the hardware parameter co-optimization model, start iterative calculation, output intermediate parameter combinations every 20 iterations, and finally generate 10 complete hardware parameter co-adjustment schemes. Based on the priority of sensor applications, the evaluation dimensions of the analytic hierarchy process (AHP) were determined, and the weight allocation rules of 40% for immunity, 35% for accuracy, and 25% for sensitivity were clarified to build a solution evaluation framework. Substitute the 10 adjustment schemes into the evaluation framework, calculate the individual scores and comprehensive weighted scores of each group in the dimensions of anti-interference, accuracy, and sensitivity, and generate a scheme score ranking table; Based on the score ranking table, the top 3 solutions were selected, and the parameters of each solution were checked to see if they met the hardware specification limits. Finally, 3 candidate adjustment solutions were determined.

7. The debugging method for the AI ​​human body sensor according to claim 1, characterized in that, The S5 includes: S51. Repeat S3 and S4 until the sensor performance reaches the preset standard and generate the final hardware parameter optimization data. Based on the final hardware parameter optimization data, sort out the core links of batch debugging. Write an automated debugging script in Python, integrate the sensor batch communication interface, environmental scene switching instructions, and performance index calculation functions to generate an automated batch debugging process scheme. S52. Select 10 sensors for small-batch trial debugging to verify the script's parameter loading success rate and scene switching stability; expand the debugging scene library and optimize the script's fault tolerance mechanism based on common production line problems. S53. Deploy the optimized automated batch debugging process to the production line debugging system. Through the linkage interface between the dedicated debugging device and the production line, realize the automated connection of sensor loading, debugging and unloading. Perform batch parameter configuration and multi-scenario debugging for large-scale production line sensors, record the debugging data of each sensor in real time, and generate a batch debugging completion dataset. S54. A stratified sampling inspection model of initial inspection + re-inspection is adopted. The initial inspection extracts 5% of the samples from the batch data and tests the core indicators in the production line self-inspection environment. The re-inspection extracts 30% of the samples that passed the initial inspection and conducts full-scenario repeated testing in an independent third-party debugging environment to verify performance consistency. S55. Summarize the batch debugging data and sampling results to generate the final debugging qualification report; and enter the debugging data of each sensor into the blockchain system.

8. An adjustment device for an AI human body sensor, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the debugging method of the AI ​​human body sensor as described in any one of claims 1 to 7.

9. An AI human body sensor, on which a computer program is stored, characterized in that, The program is executed by the processor to implement the debugging method of the AI ​​human body sensor as described in any one of claims 1 to 7.

Images