Detection instrument field intelligent calibration method and system based on camera assistance, storage medium and program product

Abstract

The invention relates to the technical field of instrument and equipment calibration, in particular to a detection instrument field intelligent calibration method and system based on camera assistance, a storage medium and a program product. Multi-view video streams are collected through a camera array, instrument readings are analyzed by using a convolutional neural network-OCR model, drift compensation is performed in combination with environmental sensor data and an LSTM model, and calibration errors are generated in real time. And the system automatically performs on-chain evidence storage on the whole-process data, and prompts a correction instruction through the AR terminal under an abnormal condition. And finally, a structured calibration report containing maintenance suggestions and block chain hash is generated, and it is ensured that the calibration result is credible, traceable and automatic.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of instrument calibration technology, and in particular to a camera-assisted detection instrument field calibration method and system, a storage medium and a program product. BACKGROUND

[0002] Detection instruments (such as pressure, temperature, flow, spectrum, and component analysis equipment) play a core quality benchmark role in high-sensitive industries such as pharmaceuticals, energy, food, and semiconductor manufacturing. More and more enterprises are required to complete periodic calibration on-site to avoid downtime losses and secondary errors caused by disassembly and transportation. However, traditional on-site calibration still relies heavily on manual operation and paper / Excel records: calibration personnel manually transcribe readings, take photos for record, fill out reports, and then return to the office to enter the system. The process is complex, subjective, and error-prone, and the calibration traceability chain lacks objective evidence support, making it difficult to meet the comprehensive requirements of real-time, traceability, and non-repudiation for digital quality management.

[0003] The current mainstream approach is to generate a unique two-dimensional code for each instrument in the measurement management platform, combined with online commissioning and certificate download functions of the verification institution, to realize electronic file management. For example, Chinese patent CN106934630A proposes a "one object one code" traceability method for instrument basic information and verification conclusions by associating a two-dimensional code label with a cloud database. This solution improves the efficiency of certificate circulation, but still relies on manual entry of results, lacks "process-level" supervision of the calibration process itself, and cannot prevent reading transcription errors or fraud.

[0004] With the decline in the price of industrial vision hardware, some enterprises have begun to install cameras on site to take photos as evidence for key steps and manually compare instrument readings. However, most systems only provide "after-the-fact review" and do not achieve automatic data extraction. Chinese patent CN105122303A discusses "camera calibration using feature recognition" technology, which extracts image features to calculate camera parameters to ensure imaging accuracy; Chinese patent CN117119113B further proposes completing camera self-calibration with user assistance to improve OCR recognition accuracy. However, both focus on camera self-calibration and do not address the industry pain point of "multi-viewpoint synchronous collection and real-time analysis of standard instrument / calibrated instrument readings."

[0005] Currently, some manufacturers are trying to introduce OCR into on-site calibration: by using an industrial camera to capture the calibration table header and using a general character recognition algorithm to extract numbers. However, in strong backlight, glare, and irregular font scenarios, the misrecognition rate is significantly higher; and a single camera angle is easily blocked by personnel. While multi-camera complementary collection can increase the success rate, it requires complex external parameter calibration and time synchronization, otherwise it is difficult to achieve millisecond-level data alignment, affecting the correctness of the one-to-one correspondence between readings and standard values.

[0006] Artificial intelligence has been widely used in the field of security monitoring for behavior recognition such as standing posture, falling, and intrusion. Based on YOLOv5 and PoseEstimation, the "AI action abnormality analysis and early warning system" can judge the compliance of personnel actions in real time and detect illegal wearing or boundary crossing behavior. The multi-modal large model has achieved more than 80 points in the accident early warning score in high-risk industries such as civil explosives and chemical industry. However, public data shows that these models are mostly oriented towards safety production and focus on abnormal behavior itself, and have not yet provided a complete logical chain closure for whether the scale-instrument reading is correctly obtained and whether the steps are executed in the order of the verification procedure.

[0007] Blockchain naturally has tamper-proof and traceable features and has been used in judicial video, agricultural traceability, and logistics signing. Chinese invention patent CN116150234A discloses a "data storage method based on blockchain", which can write the target data and hash into the chain after the user pays and return the storage certificate. In the field of measurement, some discussions propose to write the PDF certificate hash into the chain, but the large amount of video streams, environmental parameters, and real-time difference data generated in the process are often restricted by "on-chain storage" or "only hash is retained, lack of online verification entry", etc. It is difficult to truly form a strong evidence chain.

[0008] Therefore, some instrument suppliers in Europe and the United States have provided remote optimization and calibration services, such as SCIEX's "Remote Instrument Optimization and Calibration", which allows technical engineers to participate in online quality control. At the same time, AR Remote Assistance technology combines real-time video streams and virtual labeling, allowing remote experts to guide on-site operators to complete complex maintenance or debugging. However, according to public cases, AR is mostly used in maintenance or installation scenarios, and there are no existing plug-ins for error calculation specific to measurement and calibration, standardizer-instrument synchronization comparison, and the problems of unstable industrial wireless network, privacy compliance, and data storage have not been solved.

[0009] At the same time, in laboratory measurement, temperature and humidity, atmospheric pressure, magnetic field, and even vibration can cause significant drift in high-precision instruments. The traditional approach is to place a temperature and humidity meter and a barometer on site, and the calibration personnel records the environmental values and manually calculates the correction coefficient. However, in outdoor or high-temperature workshops, the environment fluctuates dramatically, and it is difficult to obtain a continuous compensation curve relying on "point measurement". Although the Internet of Things micro-sensor network (such as LoRa-WAN collection nodes) can provide second-level data, it lacks real-time fusion algorithms with calibration error models, resulting in compensation strategies still remaining in the "offline regression" stage.

[0010] In summary, although the prior art has made progress in image recognition, AI behavior analysis, blockchain evidence storage, remote collaboration, and environmental Internet of Things, there is still a lack of a systematic solution that integrates multi-camera automatic reading, AI anomaly discrimination, real-time environmental compensation, and on-chain evidence storage. In particular, in the field of detector calibration, which requires high precision, timing, and evidence chain integrity, there is still a significant technical gap. SUMMARY

[0011] The present application aims to address the core pain points in the existing detector calibration process, such as difficulty in verifying personnel identity, distortion of manual records, lack of process-level objective evidence in calibration steps, inability to compensate for environmental interference in real time, susceptibility of result data to tampering, and data siloing. It provides a camera-assisted detector calibration method that achieves precise recognition, real-time monitoring, environmental compensation, trusted evidence collection, and intelligent closed-loop management of the entire detector calibration process. Compared with existing technologies that only provide single-point functions, this method significantly improves calibration accuracy, efficiency, and traceability, representing a significant technological advancement and economic value.

[0012] To achieve the above-mentioned purposes, the present application adopts the following technical solutions:

[0013] A camera-assisted detector calibration method, comprising the following consecutive steps:

[0014] 1) Site and personnel legality confirmation:

[0015] Obtain the calibration site coordinates through GPS and / or Beidou dual-mode positioning, and compare them with the pre-stored work order coordinates. Perform unique verification of the calibration personnel's identity using face and voiceprint dual recognition.

[0016] 2) Multi-camera panoramic acquisition and calibration:

[0017] Arrange at least three high-resolution cameras with pre-set positions around the detector being calibrated. The cameras are calibrated using Zhang Zhengyou's calibration method to obtain internal and external parameters and upload them to the cloud. Start each camera simultaneously to generate multi-view video streams with timestamps.

[0018] 3) Visual-data fusion calibration:

[0019] Call the convolutional neural network-OCR combined model to automatically recognize the standard instrument readings and the values displayed by the detector being calibrated in the video stream, and write them into the local cache in a one-to-one correspondence. Calculate the difference between the real-time recognition values and the standard calibration points to obtain the first calibration error.

[0020] 3) Abnormal action and environmental comprehensive discrimination:

[0021] The human pose estimation algorithm is used to detect whether the action sequence of the calibration personnel deviates from the preset working condition; the temperature, humidity, light and vibration data are synchronously collected, the environmental drift compensation is performed on the first calibration error, and the second calibration error is obtained;

[0022] 4) Blockchain storage and AR prompt:

[0023] The image hash, environmental data and second calibration error generated in steps 2) to 4) are calculated by SHA-3-256 and written into the alliance chain, and are digitally signed by an elliptic curve P-256; if the error or abnormal score exceeds the threshold value, correction instructions are immediately displayed on the wearable AR terminal, and remote experts are allowed to mark in real time in the AR screen through WebRTC;

[0024] 5) Intelligent report generation:

[0025] Based on the differential calculation result, the first frame screenshot of the camera evidence image and the environmental drift compensation model, a structured calibration report with predictive maintenance suggestions is automatically generated, and the on-chain transaction hash is attached.

[0026] As preferred, step 2) specifically comprises the following steps:

[0027] a) At least three 4K industrial cameras are arranged around the instrument to be calibrated along more than three non-collinear points, so that their optical axes intersect to form a minimum triangular network;

[0028] b) Zhang Zhengyou calibration is performed on each camera in turn using a 9x6 checkerboard target, and the intrinsic matrix K, distortion coefficients k1, k2, p1 and p2, and the extrinsic matrix [R|t] are obtained, and the parameter file is calculated SHA-3-256 hash and uploaded to the cloud storage;

[0029] c) The IEEE-1588 precision clock is used to synchronize the time base of each camera system, so that the difference between the trigger collection times is not greater than 5ms, and the nanosecond-level timestamp is written in the frame header;

[0030] d) Each camera starts recording video at the same time under the same trigger instruction, and outputs a multi-view video stream with a unified timestamp, providing spatiotemporally consistent image data for subsequent visual-data fusion calibration.

[0031] As preferred, step 3) specifically comprises the following steps:

[0032] a) At least three cameras synchronized by IEEE-1588 clock collect I-frames with timestamps at the same trigger time;

[0033] b) Contrast-limited adaptive histogram equalization, bilateral filter denoising and YOLO-based ROI automatic cropping are performed on the I-frames in turn;

[0034] c) the cropped images are sent into the ResNet-BiLSTM-CTC convolution-OCR combined network to obtain the standard meter readings and the instrument readings to be calibrated and their confidence levels, respectively;

[0035] d) the two types of readings are written into the local cache one by one according to the time stamp, and the multi-camera results are fused through a confidence-weighted voting mechanism;

[0036] e) the fused readings are differentially calculated with the preset standard calibration points to generate the first calibration error and archive the frame hash and confidence label.

[0037] As preferred, step 4) specifically comprises the following steps:

[0038] a) the key I-frame hash, the environmental baseline hash and the second calibration error collected by the multi-camera are integrated into the evidence metadata, the SHA-3-256 digest is calculated and the digital signature is generated using the ECDSA-P256 private key;

[0039] b) the metadata are packaged into a transaction structure together with the IPFS content addresses of the video and the sensor raw files using the "consortium chain + IPFS" hierarchical strategy, and are written into the blockchain through BFT-DPoS consensus with a block time of ≤5 seconds;

[0040] c) when the compensation error or the comprehensive abnormal score is higher than the preset threshold of 0.1% FS or 0.4 points, a red correction instruction is immediately popped up on the wearable AR terminal, and the remote expert is allowed to draw vector annotations in real time in the AR view through the WebRTC channel, and the on-site operator can continue the subsequent calibration process after completing the correction according to the annotations.

[0041] As preferred, the AR terminal automatically superimposes the real-time differential results of step C on the screen area of the instrument to be calibrated based on the SLAM technology.

[0042] As preferred, when the error is above the threshold, the system automatically sends a freezing instruction to the enterprise ERP to lock the current calibration work order and prevent the instrument from being put into use; the structured calibration report contains a calibration credibility index, which is obtained by weighting the abnormal action score, the environmental steady-state score and the error residual score.

[0043] Further, the present application also provides a detection instrument field intelligent calibration system for implementing the method, comprising:

[0044] a) a positioning and identity verification unit, including a GPS / Beidou module and a face-voiceprint integrated identification terminal;

[0045] b) a camera array unit, composed of at least three 4K cameras, with hardware synchronization and self-correction functions;

[0046] c) a visual-data fusion engine deployed on the GPU edge gateway for performing OCR recognition, difference calculation and pose analysis;

[0047] d) an environment perception and compensation module including temperature and humidity, light, vibration sensors and an LSTM-based drift compensation submodule;

[0048] e) a blockchain client module for writing image hashes, compensation data and calibration errors, and performing digital signatures;

[0049] f) an AR collaboration module supporting on-site visual guidance and real-time labeling by remote experts;

[0050] g) an intelligent report generation and ERP interface module for outputting structured reports and interacting with enterprise information systems.

[0051] As a preference, the camera array unit supports H.265 encoding and RTSP streaming, and has low-light imaging capability in scenes with ≤10 lux; the visual-data fusion engine supports 60FPS real-time inference with a delay of ≤200ms; the temperature measurement accuracy of the environment perception and compensation module is ±0.1℃, and the relative humidity measurement accuracy is ±1%RH; the blockchain client module uses the BFT-DPoS consensus algorithm, with an average block time of ≤5s; the AR collaboration module is compatible with the OpenXR standard and can run on mainstream AR glasses and smartphones.

[0052] Further, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, causes the processor to implement the steps of the method.

[0053] Further, the present application also provides a computer program product comprising program instructions, which, when executed on an electronic device, cause the electronic device to perform the method.

[0054] The present application has the following technical effects due to the adoption of the above technical solutions:

[0055] 1. Full-process identity and site dual empowerment: Through real-time positioning by GPS / Beidou and face-voiceprint dual-factor authentication, the calibration personnel are one-to-one bound with work orders and on-site coordinates, completely eliminating off-site work and unauthorized personnel calibration, significantly improving the compliance and traceability of on-site calibration activities.

[0056] 2. Zero transcription error automatic collection of readings: After unified calibration, the multi-view camera array synchronously collects the display values of the standard device and the instrument under calibration; the convolution-OCR joint network completes digital analysis and mapping in milliseconds, realizes automatic difference calculation of instrument-standard device readings "one-to-one", and eliminates manual transcription and secondary entry errors.

[0057] 3. Multi-modal AI abnormal behavior instant warning: combined with human posture estimation and speech context analysis, the system can identify rule-breaking actions, irrelevant conversations, or illegal device access within <200ms, and push correction instructions to the AR terminal in real time, reducing the on-site violation rate by more than 70%, significantly ensuring the standardization of the calibration process.

[0058] 4. Real-time environmental drift compensation and uncertainty compression: through the LoRa-WAN micro-sensing network, continuously monitor multi-dimensional environmental factors such as temperature, humidity, vibration, and electromagnetic noise, and cooperate with the LSTM drift model to correct calibration errors online, so that high-precision instruments can reduce their uncertainty to laboratory level (10^-6 order) in complex on-site environments, and the error convergence speed is improved by 2 times.

[0059] 5. Non-repudiable on-chain evidence and data integrity: using a "consortium chain + IPFS fragmentation" hierarchical storage strategy, automatically write image hashes, environmental data, and calibration results to the chain and encrypt them with elliptic curve digital signatures; third parties can verify data integrity at any time, effectively preventing post-tampering, enhancing legal effectiveness and customer trust.

[0060] In summary, the present application has achieved systematic technical breakthroughs in identity verification, reading accuracy, abnormal warning, environmental compensation, data evidence, remote collaboration, report efficiency, and security compliance, etc. dimensions, overall improving the precision, efficiency and credibility of on-site calibration of detection instruments, with significant technical progress and industry promotion value. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 is the overall structure block diagram of a detection instrument on-site auxiliary monitoring and calibration system of the present application.

[0062] Figure 2 is a camera-assisted detection instrument on-site intelligent calibration method flow chart.

[0063] Figure 3 is the functional structure diagram of the camera-assisted acquisition and image recognition module in the system of the present application.

[0064] Figure 4 is the processing flow chart of the AI abnormal behavior recognition module in the present application.

[0065] Figure 5 is the model structure and data flow diagram of the environmental factor drift compensation module in the present application.

[0066] Figure 6 is the specific flow chart of the on-chain evidence module in the present application.

[0067] Figure 7 is the result binding flow chart based on OCR and automatic analysis of calibration errors in the present application. Detailed Implementation

[0068] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0069] like Figure 1 , Figure 2 As shown, a camera-assisted intelligent on-site calibration method for testing instruments includes the following sequential steps:

[0070] 1) Verification of site and personnel legitimacy: The coordinates of the calibration site are obtained through GPS / BeiDou dual-mode positioning and compared with the coordinates of the pre-stored work order; and the identity of the calibration personnel is uniquely verified by dual facial and voiceprint recognition.

[0071] 2) Multi-camera panoramic acquisition and calibration: At least three high-resolution cameras with preset positions are arranged around the instrument being calibrated. The internal and external parameters of the cameras are obtained by Zhang Zhengyou's calibration method and uploaded to the cloud. Each camera is started synchronously to generate a multi-view video stream with timestamps.

[0072] 3) Visual-data fusion calibration: The convolutional neural network-OCR combined model is called to automatically identify the standard instrument reading and the value displayed by the instrument being calibrated in the video stream, and write them into the local cache according to the one-to-one correspondence; the real-time identified value and the standard calibration point are differentially calculated to obtain the first calibration error;

[0073] 3) Comprehensive judgment of abnormal actions and environment: The human posture estimation algorithm is used to detect whether the action sequence of the calibration personnel deviates from the preset working conditions; temperature, humidity, light and vibration data are collected simultaneously, and environmental drift compensation is performed on the first calibration error to obtain the second calibration error;

[0074] 4) Blockchain Evidence Storage and AR Prompts: The image hash, environmental data and second calibration error generated in steps 2)-4) are calculated using SHA-3-256 and written into the consortium blockchain, and digitally signed by elliptic curve P-256; if the error or abnormal score exceeds the threshold, the correction instruction is immediately displayed on the wearable AR terminal, and remote experts are allowed to annotate in the AR screen in real time via WebRTC.

[0075] 5) Intelligent report generation: Based on the differential calculation results, the first frame screenshot of the camera evidence image and the environmental drift compensation model, a structured calibration report with predictive maintenance suggestions is automatically generated and an on-chain transaction hash is attached.

[0076] The following is described in accordance with the "preparation - identity and site authority - multi-camera calibration - environmental factor baseline establishment - calibration collection and error calculation - AI anomaly discrimination - real-time compensation for environmental drift - on-chain notarization - AR collaboration - report generation - data archiving and maintenance" eleven stages, thirty-two steps

[0077] The first stage of preparation

[0078] 1.1 Hardware layout planning

[0079] The site supervisor first determines the installation position of the camera array and environmental sensing node according to the model, installation height and operation channel width of the instrument to be calibrated (hereinafter referred to as "calibrated instrument"). In principle, more than three cameras should be able to form a minimum triangular network to obtain complete complementary view angles. If the calibrated instrument has more than two display panels, additional cameras can be installed in front of the corresponding panels. All cameras and sensing nodes are fixed using industrial magnetic bases or expansion bolts to meet the requirements of shockproof and IP65 protection level.

[0080] 1.2 Power on and network connection check on edge gateway

[0081] The GPU-edge-gateway (referred to as "edge gateway") is fixed on the wall cabinet within 5m of the instrument. The edge gateway is connected to the enterprise workshop backbone network through a gigabit Ethernet and is configured with a dual-frequency Wi-Fi 6 AP to provide a wireless backup link for cameras, AR glasses and environmental nodes. The technician connects the 220V industrial power supply and checks that the UPS module self-test is passed before proceeding to the next stage.

[0082] The second stage of identity and site authority

[0083] 2.1 Upload of work order geographic coordinates

[0084] The calibration scheduling system automatically writes the latitude and longitude coordinates (WGS-84) when the invoice work order is issued, and pushes it to the PostgreSQL database of the edge gateway. The error threshold for comparing the real-time readings of the GPS / Beidou module with this coordinate is set to 10m; exceeding the threshold will trigger a red warning and prevent calibration from starting.

[0085] 2.2 Personnel identity template issuance

[0086] The enterprise digital identity platform calls the OAuth2.0 interface to issue the face feature vector, voiceprint feature vector and employee number of the authorized personnel for this calibration to the local TEE of the edge gateway; at the same time, a one-time token is generated for the calibration personnel to scan and log in on the AR glasses side. This token is invalid after 10 minutes to prevent it from being intercepted and replayed.

[0087] 2.3 On-site check-in and authority

[0088] The calibration personnel arrives at the scene, and through the AR glasses, the face is directly detected and the face vector is compared. Then the system randomly generates a five-digit verification code, and the voiceprint verification is completed. The system signs and chains the double-factor verification result together with the real-time GPS coordinates. If any link fails, the calibration process is automatically terminated and recorded as an "illegal attempt" event.

[0089] Third stage multi-camera calibration

[0090] Step 3.1 External parameter measurement and rough calibration

[0091] Use a laser range finder to obtain the distance and angle between the camera and the center point of the surface of the instrument being calibrated. Input the range data into the calibration assistant program to automatically generate the initial external parameter matrix E0 for each camera. This matrix includes a rotation matrix R ∈ R3×3 and a translation vector t ∈ R3×1.

[0092] Step 3.2 Chessboard fine calibration

[0093] Place a 9x6 chessboard calibration target next to the instrument being calibrated. The system collects 20 frames of static images and uses Zhang Zhengyou's algorithm to obtain the intrinsic matrix K and distortion coefficients (k1, k2, p1, p2). Generate a calibration report file and calculate the SHA-3-256 hash value as Hcamcal.

[0094] Step 3.3 Time synchronization marking

[0095] To ensure that the multi-camera frames are aligned, start IEEE1588 PTP (Precision Time Protocol) on the edge gateway to synchronize the camera system clock to the gateway's local crystal oscillator. Finally, verify that the frame synchronization error is less than 5ms by testing the signal light blinking.

[0096] Fourth stage environmental factor baseline establishment

[0097] The goal of this stage is to establish a dynamic baseline for key environmental quantities such as temperature, humidity, vibration, electromagnetic noise, air pressure, and illumination within 15 minutes before calibration begins, so that the subsequent compensation model can accurately correct transient environmental drift. The implementation steps are divided into six sub-processes: hardware initialization, sensor node self-calibration, wireless networking, baseline collection, statistical integration, and model hot loading.

[0098] 4.1 Hardware initialization

[0099] (1) Node type and distribution

[0100] Each system is equipped with 6 types of LoRa-WAN micro-sensing nodes by default: temperature-humidity integrated node TH-01, three-axis acceleration node V-03, wideband electromagnetic noise node EM-02, absolute pressure node P-01, light node LUX-01, and backup AI camera node VIS-B (for image correlation).

[0101] Node arrangement principle: TH-01, P-01 are installed 20 cm away from the sensing panel of the instrument being calibrated; V-03 is attached to the base plate of the machine stand; EM-02 is placed on the top of the cabinet; LUX-01 is suspended 1 m above the horizontal center of the instrument. The straight-line distance from the node to the center of the instrument being calibrated is not more than 1 m, ensuring that the environmental quantity is representative.

[0102] (2) Power-on self-test

[0103] After the node is powered on, the MCUBoot-ROM built-in BIST (Built-In Self-Test) sequence is executed first to quickly scan the sensing chip, power stability, and RTC clock drift.

[0104] If the self-test result flag contains any bit error (for example, the temperature ADC check value does not match), the node will show a warning through the local LED and refuse to join the LoRa network.

[0105] 4.2 Self-calibration of sensing nodes

[0106] (1) Temperature-humidity integrated node

[0107] The node internally stores the last multi-point correction coefficient C T,i ,C RH,i . During self-test, the on-chip 100kΩ precision reference thermistor level is read and compared with the calibration curve in the EEPROM; if the deviation exceeds ±0.05℃, a second-order polynomial compensation is enabled:

[0108]

[0109] (2) Three-axis vibration node

[0110] The static zero offset of the internal MEMS accelerometer is estimated b x ,b y ,b z The gravity vector module is calculated, and if Automatic zero offset reset is performed and the event is recorded.

[0111] (3) Electromagnetic noise node

[0112] Support 9kHz-30MHz wideband FFT, self-calibration phase on 100kHz fixed signal generator for phase-locked scanning, verify that the passband gain error is not more than 0.5dB; if it exceeds, the node switches to the backup gain table.

[0113] (4) Air pressure and light node

[0114] The air pressure node reads the factory sealed vacuum cavity comparison value; the light node uses the reference response triggered by the on-chip 550nm LED to do a slope calibration, ensuring that the linear error is <1%.

[0115] After the self-calibration is completed, all the correction coefficients are written into the node Shadow RAM, and a 32-bit CRC check code is generated and reported.

[0116] 4.3 LoRa-WAN wireless networking

[0117] (1) Join-Procedure

[0118] The node uses the OTAA (Over-The-Air-Activation) method to join the network, and attaches the previous CRC when sending the Join-Request; the edge gateway LoRa Server quickly checks the CRC.

[0119] After the server returns the Join-Accept, the node automatically switches to the latest Adaptive Data Rate, ensuring 250bps in an empty workshop and 980bps in a metal-intensive workshop.

[0120] (2) Time synchronization

[0121] The LoRa-WAN Class B downlink Beacon frame is transmitted once every 128s, and the node aligns with the BeaconCounter using the temperature-compensated crystal oscillator to achieve a time base consistency of ±20ppm, providing a unified time axis for subsequent sliding window averaging.

[0122] 4.4 Baseline data acquisition

[0123] (1) Sampling frequency

[0124] TH-01: 2s once; V-03: 1s once; EM-02: 5s once; P-01 and LUX-01: 10s once. Each sample contains the measured value and Node-TX-Timestamp (μs), and when it arrives at the gateway via LoRa-WAN, it is accompanied by the gateway RX-Timestamp.

[0125] (2) Edge gateway cache

[0126] The gateway writes data into the ring buffer and sorts by Node ID, forming the following time series

[0127] {T i (t k )},{RH i (t k )},{V RMS，i (t k )},{N PSD，i (t k )},{P i (t k )},{L i (t k )};

[0128] where i is the node number, t k is the sampling time.

[0129] (3) Outlier rejection

[0130] For each series, perform a three-point sliding median filter; if a single point deviates from the average of its neighbors by more than 3σ, mark it as an instantaneous outlier and record the Node ID and deviation in the log, then replace it with a linear interpolation.

[0131] 4.5 Statistical integration

[0132] (1) Sliding window mean and variance calculation

[0133] Select a 10 min window (≈300 TH-01 sampling points) and calculate the mean and standard deviation σ T , σ RH , … of each environmental quantity. If σ T >0.3℃ or σ RH >2, etc., consider the environment unstable, and the system issues a yellow warning on the AR glasses and extends the baseline collection by 5 min.

[0134] (2) Baseline vector generation

[0135] When the standard deviations of all environmental quantities fall within the threshold, the edge gateway calculates the baseline vector

[0136]

[0137] and stores it in the local Influx-DB.

[0138] (3) Hash digest

[0139] Pack all original sampling sequences in Message Pack and calculate the SHA-3-256 hash to obtain H env_raw ; then calculate the E baseThe hash H is calculated by packing the standard deviation array together. env_bas e. Both are written into the blockchain along with the camera's calibration hash.

[0140] 4.6 Model Hot Loading and Drift Threshold Setting

[0141] (1) Loading the LSTM model

[0142] Based on the current instrument model and range being calibrated, the gateway retrieves the latest version number v from the model repository. n LSTM drift model weight file Θ vn The weights are quantized in 16-bit INT8 format. After loading, the ECDSA-P256 signature of the file is compared with the SHA-3-256 checksum; only when both pass the checksum are the data entered into the GPU inference RAM.

[0143] (2) Temperature window threshold matching

[0144] in accordance with The temperature range is determined by referring to a table to obtain the temperature sensitivity coefficient K. T And add a linear gain before the model inference layer:

[0145]

[0146] (3) Quick drift valve

[0147] To avoid excessive calculation deviations during the LSTM cold start phase, a fast drift valve is implemented:

[0148] |δ env (t)|≤C·σ Δ,spec ;

[0149] Where C = 0.5–0.7, σ Δ,spec The maximum permissible standard deviation given by the instrument specifications. If the predicted value exceeds the threshold, it is replaced with the threshold value and marked "Estimation Uncertain".

[0150] Through the six sub-processes described above, this invention completes self-calibration, stability assessment, and baseline determination for multiple environmental parameters within 15 minutes, ensuring that the subsequent LSTM drift model can be built on reliable and highly consistent input data. The precise implementation of this stage directly determines the compensation accuracy, thus affecting the overall calibration uncertainty. Actual measurements show that even in the harsh environment of a metal heat treatment workshop (temperature fluctuation ±2℃, humidity fluctuation ±6%RH), this solution can still control the calibration residual after drift compensation within 110% of the laboratory-level uncertainty, fully verifying the effectiveness and robustness of this stage of the process.

[0151] Phase 5 Calibration Data Acquisition and Error Calculation

[0152] This stage is the core measurement process that officially begins after the environmental baseline is established, camera calibration is completed, and personnel identification is confirmed. It directly determines the metrological traceability validity of the calibration results. To enable those skilled in the art to replicate this process in different testing projects, this implementation method subdivides the fifth stage into twelve steps. The following description uses a 0.1% class pressure transmitter (range 0–1 MPa) as the instrument being calibrated and a 0.02% class pressure control / measurement module as the standard; however, except for the range and unit, all steps can be seamlessly replaced with other physical quantities such as temperature, flow rate, pH, and conductivity.

[0153] 5.1 Steady-state verification of the standard instrument

[0154] (1) Steady-state start command

[0155] The edge gateway sends a steady-state preparation command to the pressure control module via the SCPI instruction INIT:PROCSTABLE. The standard's internal closed-loop feedback PID control loop locks the output pressure to the current calibration point P. set .

[0156] (2) Steady-state criterion

[0157] The gateway reads the standard's digital interface return value P at 5Hz. std (t). If in consecutive n... s = Satisfy within 30 seconds

[0158]

[0159] Then it is determined that a steady state has been reached; where ∈ s =0.01%FS.

[0160] (3) Temperature Co-check

[0161] Simultaneously, it was verified that the temperature node acquisition value T(t) satisfies: |T(t)-T(t-30s)|<ε T Otherwise, the process will be delayed until the next step is reached and a message will be displayed indicating "waiting for steady state".

[0162] 5.2 Synchronous Triggering and Time Base Locking

[0163] (1) PTP clock fine alignment

[0164] The edge gateway acts as the PTPPGrandmaster, broadcasting nanosecond-level timestamps to all camera, microphone array, and standard interface log threads; the camera's internal PLL phase-locked loop reports SYNC_OK.

[0165] (2) Single-trigger design

[0166] When steady-state condition is met, gateway sends out TRIG multicast packet. Camera array receives TRIG and captures an I-frame at the same PTP time t0 and returns frame number and timestamp immediately; if the difference is >5ms, record ASYNC_WARN and enter the re-shooting branch (see 5.9).

[0167] 5.3 Image ROI adaptive cropping

[0168] (1) Panorama to local

[0169] Each camera transmits the 4K raw frame to GPU—edge—gateway; gateway runs YOLOv8-Det to detect the candidate bounding box B of the instrument display window and the standard display window dev , B std .

[0170] (2) Parallax correction

[0171] According to the camera extrinsic matrix R, t, project the bounding box to the common reference plane; if the center of the projected box deviates from the last frame by >10px, it is considered that the lens may be shaken and needs to be automatically adjusted or prompted for manual reset.

[0172] 5.4 Convolution-OCR reading extraction

[0173] (1) Feature pyramid enhancement

[0174] The cropped ROI picture is enhanced by CLAHE (Contrast Limited Adaptive Histogram Equalization) and sent to CRNN-OCR. A lightweight PP-OCR-v4Head is used for character sequence recognition, and the result string_dev and confidence p_dev are returned; similarly, string_std and p_std are obtained.

[0175] (2) Character post-processing

[0176] Regular rules ^\d{1,3}(\.\d{1,4})?$ are executed to ensure the legality of the value; if the confidence p is <0.85 due to intermittent character reflection, call the cosine similarity backup language model LM_corr for correction prediction.

[0177] 5.5 Multi-camera result consistency arbitration

[0178] (1) Voting-confidence mechanism

[0179] The reading results of the three cameras are recorded as {P dev ,j},{P std, j = 1, 2, 3. If two results are consistent and both have confidence > 0.9, take their arithmetic mean; if all three are inconsistent, take the mode; if still no mode, choose the single one with the highest confidence and remark SINGLE_RELY.

[0180] (2) View angle weight adaptation

[0181] System maintains a camera reliability weight table w j If a camera is judged as low confidence or SINGLE_RELY for 5 times in a row, its weight w j is decreased by 0.1, with the minimum of 0.2; the weight is used for subsequent weighted average

[0182]

[0183] 5.6 Digital interface cross-checking

[0184] (1) Standard redundancy comparison

[0185] In addition to image OCR, the gateway retrieves the internal A / D reading of the standard through RS-485

[0186] Compare P std from OCR with P from A / D If the difference is greater than 0.5% of the standard's full scale, alarm and request repeated shooting to avoid partial screen glare leading to misreading.

[0187]

[0188] Alarm and request repeated shooting to avoid partial screen glare leading to misreading.

[0189] 5.7 Differential error calculation

[0190] (1) Definition

[0191] Original differential: ΔP1 = P dev -P std , unit: Pa.

[0192] (2) Range normalization

[0193] To facilitate comparison of devices with different ranges, calculate the normalized error:

[0194]

[0195] 5.8 Repeated measurement and statistical filtering

[0196] (1) Number of cycles setting

[0197] Collect nr = 5 repeated measurements for each calibration point, with an interval of 3s; the collection sequence is denoted as

[0198] (2) Outlier rejection rule

[0199] Calculate average with standard deviation σ δ ; if exists

[0200]

[0201] then it is an outlier and marked OUTLIER_k in the log. Outliers will be rejected before final average is calculated

[0202]

[0203] where n r * is the number of valid samples (> 3).

[0204] 5.9 Retake and rollback mechanism

[0205] (1) Frame desynchronization remedy

[0206] If ASYNC_WARN is detected in 5.2, the gateway will immediately issue a secondary TRIG and record RETAKE_FLAG = 1 in the metadata; only the latest I-frame that successfully synchronized will be saved.

[0207] (2) OCR failure rollback

[0208] If the OCR confidence of a camera is < 0.7 for three consecutive times, the "angle suggestion algorithm" will be started: calculate the best possible shooting direction according to the extrinsic matrix and prompt the operator to adjust through the AR arrow.

[0209] 5.10 Real-time visual feedback on site

[0210] (1) AR data floating window

[0211] AR glasses HUD presents P std , P dev and δ1 * in real time, and uses color to distinguish (green: |δ| < 0.05% FS; yellow: 0.05-0.1%; red: > 0.1%). If red appears, the system will pause and enter 5.11, allowing the operator to choose "repeat / confirm fault".

[0212] (2) Voice broadcast

[0213] The Text-to-Speech kernel will broadcast "Calibration point X successful, error Yppm" after successful sampling, improving the efficiency of on-site operations.

[0214] 5.11 Calibration point sequence management

[0215] (1) Automatic step pressure control

[0216] Switch the standard pressure automatically with P_set_next = P_set_prev + ΔP_step, ΔP_step is set to 20% FS according to the regulation. The gateway maintains the state machine STATE = {IDLE, STABILIZING, CAPTURING, COMPLETE} and updates the ERP work order in real time.

[0217] (2) Bidirectional return inspection

[0218] After completing the rising step, automatically decrease the same step, compare the return error

[0219]

[0220] If δ return Exceeds 0.02% FS, it indicates the risk of mechanical hysteresis, and reports the "return deviation" event.

[0221] 5.12 Data packet and temporary storage

[0222] (1) Sampling packet format

[0223] Write into the edge gateway SSD after compression with MessagePack, and generate the signature sig_step5.

[0224] (2) Temporary backup and confirmation

[0225] After completing each calibration point, the SSD will push the encrypted copy to the nearest computer room NAS through SFTP; after the NAS returns ACK, the gateway will cut into the next calibration point.

[0226] Through the twelve-step detailed process, the present application completes high-trust collection and difference calculation of a calibration point under the chain logic of standardizer steady state - synchronous trigger - multi-source consistency - repeated statistics - instant feedback - sequence management - packet signature. Compared with the traditional hand-copying - comparison process, OCR and digital bus double redundancy, outlier inspection and bidirectional return inspection greatly reduce random error and systematic deviation, so that the normalized error residual can still be better than 0.05% FS under industrial conditions; at the same time, the synchronous frame mechanism ensures that the image, audio and environmental data form the same time domain digital evidence, which lays a foundation for subsequent on-chain notarization and automatic reporting.

[0227] Sixth stage AI abnormal behavior identification

[0228] The core purpose of this stage is to identify and judge the operation behavior, voice content and equipment state in the on-site calibration process of detection instruments through multi-modal artificial intelligence models in real time, prevent the occurrence of illegal behaviors such as operator substitution, non-procedure operation, non-standard equipment access, cheating negotiation, and thus improve the credibility and irrefutability of the entire calibration process.

[0229] 6.1 Video stream input and pose skeleton extraction

[0230] (1) Synchronous video stream capture

[0231] A multi-camera array (≥3) starts recording when each calibration action point (such as turning on the standard device, reading the display, connecting the pipeline) is triggered, with a frame rate of 30fps. The video stream data is uniformly transcoded into RGB image sequence format, and clock synchronization is performed through PTP (Precision Time Protocol) to ensure subsequent multi-modal alignment.

[0232] (2) Human pose estimation network processing

[0233] Each frame of image input is processed by the YOLOv8-Pose model, and the two-dimensional coordinates of 17 key points of the human body are output:

[0234] P i (t) = (x i (t), y i (t)), i ∈ {1, 2,..., 17},

[0235] where P i (t) represents the pixel coordinates of the i-th body key point (such as right shoulder, left wrist) at time t.

[0236] (3) Skeleton trajectory time series construction

[0237] The 17 key points are connected as pose trajectory vector sequences on consecutive time frames:

[0238]

[0239] and a 10-second sliding time window data segment is constructed for behavior recognition.

[0240]

[0241] (1) Action classification network loading

[0242] The above sliding window pose sequence is input into a lightweight action recognition network (such as the ST-GCN or MoveNet-LSTM hybrid model), and the current behavior label A(t) and confidence p A(t). Model categories include but are not limited to:

[0243] Compliant behavior: holding device, key confirmation, device alignment, observing display;

[0244] Abnormal behavior: long absence, body blocking camera, facing others, using non-standard tools, etc.

[0245] (2) Abnormal action score calculation

[0246] According to the action label and confidence, set the abnormal action score function:

[0247]

[0248] If S pose (t) > 0.3 for more than 3 times in a continuous time window (such as 5 seconds), an "action abnormality" event is triggered.

[0249] (3) Multi-person operation discrimination

[0250] The system detects whether other personnel have entered the operation area or identifies unauthorized personnel operating the calibrated instrument through the number of skeletons between frames , and immediately prompts "unauthorized personnel intervention" in the AR glasses as soon as it is found.

[0251] 6.3 Voice monitoring and semantic anomaly recognition

[0252] (1) Microphone array and sound source positioning

[0253] A microphone array composed of 4 high-sensitivity directional microphones is used for beamforming to lock the direction of the main voice source, shield background noise and irrelevant conversations. Extract a 10-second audio segment and send it to the speech recognition system.

[0254] (2) Speech transcription and context classification

[0255] Whisper or Conformer-RNN model is used for ASR (Automatic Speech Recognition) processing, and the speech transcription text T voice (t) is output.

[0256] The transcription text is input into a natural language classifier (based on BERT or ERNIE model) for context label judgment:

[0257] Compliant statements: reading pressure values, operation instructions, execution confirmation;

[0258] Abnormal statements: discussing non-standard operations, private coordination, irrelevant chatter, refusing supervision, etc.

[0259] (3) Semantic anomaly score

[0260] Let the current context label be L voice (t), whose corresponding anomaly score is defined as:

[0261]

[0262] If the average Label "Voice anomaly".

[0263] 6.4 Device recognition and illegal tool access detection

[0264] (1) Object recognition and label comparison

[0265] Use YOLOv8-Object-Detection to detect tools, connectors, cables, and other devices in live video frames; and import a legal device library from the ERP work order

[0266] For each frame of recognition result D t ={tool i} and the legal device set, perform a difference set comparison:

[0267]

[0268] If illegal devices (such as non-standard pipe connectors, non-brand standard devices) are detected, record the image screenshot and label "Tool anomaly".

[0269] (2) Device anomaly score calculation

[0270] Define the device anomaly score:

[0271]

[0272] That is, the proportion of the number of illegal tools, if S device >0.2 for 3 consecutive frames, trigger an alarm.

[0273] 6.5 Multi-modal comprehensive anomaly score and alarm mechanism

[0274] (1) Anomaly score fusion model

[0275] Construct a multi-modal anomaly score function:

[0276] S abn (t) = a · S pose (t) + b · S speech (t) + g · S device (t),

[0277] The weight recommendation is set as: a = 0.4, b = 0.3, g = 0.3, which can be adjusted according to the actual industry sensitivity.

[0278] (2) Window average and event trigger

[0279] Calculate the average anomaly score in a 15-second sliding window If the following conditions are met:

[0280]

[0281] Determine that the current operation has a risk of violation, and trigger the following processing:

[0282] a) Pop up a red box in the AR glasses with the prompt "abnormal operation, please recheck";

[0283] b) System voice broadcast prompt;

[0284] c) Automatically pause data collection and error recording process, and the on-site personnel need to re-execute the process;

[0285] Pack the abnormal related video, voice segment, and equipment image into the chain preparation block and write it.

[0286] 6.6 Log generation and on-chain marking

[0287] (1) Log format

[0288] The system generates an AI anomaly analysis log for each calibration point.

[0289] (2) Hash and chain writing preparation

[0290] Pack the log to generate hash H ai_log = SHA-3-256 (log), which will be written into the blockchain together with the image hash and calibration error value in the future, ensuring the non-repudiation of the abnormal processing result and the third-party auditing ability.

[0291] In this phase, by combining posture recognition, voice analysis, and equipment image recognition, a fusion scoring system is constructed to ensure comprehensive, real-time, and intelligent monitoring of operation behavior during instrument calibration. The significant technical effects include:

[0292] a) Millisecond-level response to unauthorized personnel, wrong steps, verbal negotiation cheating, etc.

[0293] b) Improve the compliance, credibility, and legal effectiveness of the calibration process;

[0294] c) Automatically form AI judgment logs to provide effective evidence for digital forensics and on-chain evidence.

[0295] The multimodal AI anomaly identification model of the application can be iteratively updated, adapt to more industry standard operation processes, and has good universality and technical scalability.

[0296] Real-time compensation for environmental drift in the seventh stage

[0297] This stage aims to solve the error problems of zero point shift, response nonlinearity or hysteresis caused by temperature, humidity, air pressure, vibration, electromagnetic interference and other factors in the field environment of the detection instrument, so as to dynamically correct the original error data and improve the objectivity and accuracy of the calibration data. The application proposes an environmental drift real-time compensation method based on long short-term memory neural network (LSTM) modeling, which combines real-time collected multi-dimensional environmental parameters, performs fast reasoning on the edge computing gateway, and outputs compensation values by fusing with current error data. The implementation process is carried out in the following seven steps:

[0298] 7.1 Environmental data stream acquisition and time sequence arrangement

[0299] 7.1.1: Multi-source synchronous acquisition

[0300] All environmental sensor nodes (temperature, humidity, air pressure, vibration, electromagnetic) deployed around the instrument to be calibrated are connected with the edge gateway through LoRa or Wi-Fi, and the sampling frequency is as follows:

[0301] Temperature, humidity: every 2s;

[0302] Air pressure, illumination: every 5s;

[0303] Three-axis vibration: every 1s RMS calculation;

[0304] Electromagnetic noise: every 10s spectrum mean.

[0305] 7.1.2: Definition of environmental vector standardization

[0306] At each time point t, a multi-dimensional environmental state vector is defined:

[0307]

[0308] Wherein:

[0309] T(t): temperature (℃);

[0310] RH(t): relative humidity (%);

[0311] P(t): air pressure (hPa);

[0312] V(t): three-axis combined vibration RMS (m / s 2 );

[0313] N(t): Mean of electromagnetic noise (dBμV).

[0314] All environmental quantities are linearly normalized before inputting into LSTM:

[0315]

[0316] where μ i ,σ i are the empirical mean and standard deviation of each environmental dimension (can be set according to the last 30 minutes sliding window statistics or laboratory historical values).

[0317] 7.2 LSTM Drift Model Architecture and Loading

[0318] 7.2.1: Model Structure Definition

[0319] The invention uses a double-layer stacked LSTM network structure to construct the environmental drift prediction model:

[0320] Input layer dimension: 5 (number of environmental variables);

[0321] First layer LSTM: hidden layer size 64;

[0322] Second layer LSTM: hidden layer size 32;

[0323] Fully connected layer: output predicted drift δ env (unit same as the calibrated quantity, such as Pa or ℃);

[0324] Time window: use 60 seconds of historical data to construct time series input, step 1s.

[0325] 7.2.2: Model Training and Hot Update Mechanism

[0326] The offline stage of the model uses the laboratory calibrated environment and error pairs for supervised training, using the MSE loss function; after online deployment, the edge gateway receives a new weight package (.pth format) every quarter for hot replacement; before weight update, integrity check (SHA-256 hash comparison + ECDSA signature verification) must be performed.

[0327] 7.3 History Time Window Construction and Input Preparation

[0328] 7.3.1: History Sequence Cache

[0329] The edge gateway maintains a one-dimensional environmental vector sequence sampled every 1 second for the past 60 seconds:

[0330]

[0331] An input tensor of dimension 60x560x560x5 is constructed and input into the LSTM model for inference.

[0332] 7.4 Drift error prediction and output

[0333] 7.4.1: LSTM inference process

[0334] The input tensor E(t) is forward propagated through the model, and the predicted environmental drift error at the current time is output:

[0335]

[0336] Where:

[0337] The trained LSTM model;

[0338] δ env (t): unit of the same physical quantity to be calibrated, such as pressure (Pa), temperature (°C), flow rate (L / min), etc.

[0339] 7.4.2: Error stability judgment

[0340] To avoid model cold start jitter, set the maximum drift threshold δ max For example:

[0341] |δ env (t)|≤δ max = 0.05% FS,

[0342] If this value is exceeded, it is determined to be an abnormal prediction, and the clipping value is used instead and marked DRIFT_CLIP_FLAG = 1.

[0343] 7.5 Error compensation and real-time update

[0344] 7.5.1: Definition of original error

[0345] The original differential error (uncompensated) obtained in the calibration data collection is:

[0346] Δ raw (t) = V dev (t) - V std (t).

[0347] 7.5.2: Error calculation after compensation

[0348] The drift amount predicted by the LSTM model is compensated to obtain the corrected differential error:

[0349] Δ corr (t) = Δ raw (t) - δ env (t),

[0350] The compensation error will be used as the official error value for subsequent on-chain attestation and report generation.

[0351] 7.6 Real-time Drift Status Indication and Alarm Mechanism

[0352] 7.6.1: Credibility Score Mechanism

[0353] Calculate the credibility score C(t) of the current drift prediction, defined as follows:

[0354]

[0355] If C(t) > 0.9, display green state (model credible);

[0356] 0.7 < C(t) ≤ 0.9, display yellow (medium trust);

[0357] C(t) ≤ 0.7, display red, trigger "drift uncertainty alarm".

[0358] 7.6.2: AR Real-time Display

[0359] In the wearable AR glasses, the system will overlay the current predicted drift value, credibility, and compensated error curve in real time on the operation interface, and provide voice prompts such as "environment compensation started / drift is high" and other information.

[0360] 7.7 Compensation Data Recording and Traceable Storage

[0361] 7.7.1: Original and Compensation Value Recording

[0362] For each sampling point, the system records: original error Δ raw (t); environmental drift value δ env (t); compensated error Δ corr (t); drift credibility score C(t).

[0363] 7.7.2: Data Packaging and Signing

[0364] Pack the above data and sampling timestamp into a JSON document, generate a SHA-3-256 hash digest, and sign it with the gateway private key using ECDSA-P256, ready to be used as part of the on-chain attestation data structure.

[0365] The present application realizes real-time modeling and quantification of the influence of complex dynamic field environment on measurement accuracy by constructing an LSTM-based environmental drift error prediction model, and applies drift compensation accurately to each calibration sampling point through an online inference mechanism, effectively improving the comparability and credibility of data. The system design has the following advantages:

[0366] a) Strong adaptability: model can be trained and migrated, suitable for different physical quantities and devices;

[0367] b) High real-time performance: GPU inference delay <100ms;

[0368] c) Intuitive visualization: combined with AR real-time prompts, operators can actively perceive drift risks;

[0369] d) Traceability: all compensation data is signed and chained, ensuring the credibility of the calibration closed-loop.

[0370] This stage is the precision guarantee core of the entire intelligent calibration system and is one of the key innovations that distinguish this invention from traditional methods.

[0371] Eighth stage: on-chain evidence

[0372] The I-frame image file of this calibration point, the standard value, the instrument value to be calibrated, the environment vector, and the abnormal score are spliced into an original byte stream, and then SHA-3-256 is used to generate a hash H pt . The ECDSA-P256 private key in the edge gateway TPM chip is used to digitally sign H pt σ TPM . Then construct the blockchain transaction:

[0373] Tx = 〈H pt , σ TPM , timestamp, nodeID >,

[0374] and broadcast to the consortium chain. The main chain confirmation time is ≤5s. The transaction index hash H tx will be cited in the report.

[0375] Ninth stage: AR real-time collaboration

[0376] If the on-site personnel need assistance, they can click the AR menu to initiate a WebRTC session and invite the measurement expert to join. The expert can see the panoramic video stitched by multiple cameras in the browser and can draw lines and points at any position on the screen. All markers are projected back to the AR field in real time. The system automatically generates 3D arrows or text bubbles based on expert markings and error feedback. For example, "rotate the adjustment knob to 1.2bar", the text floats next to the knob; "replace the calibration point to 150℃", floating above the temperature control knob. This entire interactive dialogue is recorded by websocket and generates a signed hash.

[0377] Tenth stage: report generation

[0378] The edge gateway calls Matplotlib-CPU to plot the curve of AV2(t) over time and overlay the ambient vector curve, and outputs a PNG file. The first frame of image is attached to the report as an illustration after being cropped by the function. The template engine is called to write the following fields to an XML file:

[0379] Device information: model, serial number, work order number

[0380] Calibration parameters: reference point list and corresponding AV2

[0381] Ambient baseline and instantaneous value

[0382] AI anomaly event list and timestamp

[0383] On-chain transaction hash H tx List

[0384] Calibration confidence index CI.

[0385] Where:

[0386]

[0387] Parameter k is the steepness coefficient, S abn,avg is the average anomaly score, and σ spec is the maximum allowed deviation specified by the instrument.

[0388] The generated PDF report is signed by an X.509 digital certificate and pushed to the enterprise ERP. After receiving the receipt, the ERP sets the status to "calibration completed" and the on-site device is unlocked.

[0389] Eleventh stage data archiving and maintenance

[0390] After the edge gateway completes the task, it copies the original I-frame, environmental data, OCR results, and on-chain transaction list to a RAID6-NAS, and after the on-duty engineer signs, it is transferred to an offline tape warehouse. The system performs clustering analysis on the collected environmental-error pairs every quarter and re-trains the LSTM drift model. After the new model is verified to be qualified, it is upgraded using rolling, and the old version is automatically sealed and retained for one year. According to the requirements of ISO / IEC27040, the key information log is saved for not less than ten years. The monthly execution of a blockchain transaction reconciliation, randomly selected 5% of the calibration points, the local I-frame hash is compared with the on-chain hash, and the consistency is verified, if differences are found, the security investigation process needs to be started.

[0391] Test example

[0392] I. Measured object and experimental configuration

[0393] Device under test: Model WZ-PT800 intelligent pressure transmitter, range 0-1 MPa, accuracy class 0.1% FS.

[0394] Standard: Fluke 6270A high-precision pressure controller, range 0-1 MPa, accuracy 0.02% FS.

[0395] Calibration point setting: 0%, 20%, 40%, 60%, 80%, 100% FS, a total of 6 points, rising and returning twice, a total of 12 points.

[0396] Field conditions: summer steel structure workshop (non-air conditioning), temperature fluctuation ± 2.5℃, humidity fluctuation ± 8% RH, intermittent fan and vibration interference.

[0397] The system deployment configuration is as follows:

[0398] Module Configuration Description Camera Array 3 sets of 4K industrial cameras, 120° field of view, hardware synchronization Environmental Sensing Nodes Temperature and humidity, electromagnetic noise, three-axis vibration, air pressure, illumination AR Device Rokid Glass 2, real-time feedback and expert collaboration AI Engine YOLOv8-Pose, Whisper, small BERT (edge inference) Blockchain Platform Hyperledger Fabric + IPFS (consortium chain deployment)

[0399] II. Experimental procedure

[0400] 1. Experimental control group setting

[0401] Group Description Group A (control group) Use traditional manual calibration, manual reading of values, paper records, no compensation, no video recording Group B (invention) Enable complete system: camera OCR, AI behavior recognition, environmental compensation, on-chain evidence

[0402] Each group performs 5 sets of field calibration operations, a total of 60 calibration points.

[0403] 2. Calibration implementation process

[0404] The control group is manually recorded by experienced calibrators using standard tables. The AI real-time monitoring behavior compliance, and the drift model based on LSTM real-time error offset. All B group data is automatically signed and written into the alliance chain, and is real-time connected with the ERP system.

[0405] III. Key experimental data and results

[0406] 1. Error statistical data comparison

[0407] Calibration Points Mean Error (MPa) of Control Group Mean Error (MPa) of Invention Average Difference (%) 20% FS 0.2005 0.2002 -0.15％ 40% FS 0.4007 0.4001 -0.30％ 60% FS 0.5996 0.6000 +0.07％ 80% FS 0.7992 0.8001 +0.11％ 100% FS 1.0008 1.0002 -0.06％

[0408] Conclusion: The average error deviation of the invention group after compensation is less than ± 0.1% FS, while the traditional group has a maximum of 0.3% FS offset, indicating that the environmental drift compensation model effectively reduces the temperature and humidity interference.

[0409] 2. Abnormal identification accuracy statistics

[0410]

[0411]

[0412] Conclusion: AI model has high detection ability for key violation behaviors, and can timely block potential operational risks.

[0413] 3. Data credibility and auditability

[0414] Dimensions Control Group Invention Group Error Source Traceability × √ (image + environment + error on chain) Process Compliance Record × √ (AI behavior log) Abnormal operation alarm mechanism × √ (AR pop-up + sound prompt) Digital certificate + report signature × √ (blockchain hash + signature) Report verifiability × √ (contains IPFS file pointer)

[0415] Four, result analysis

[0416] Through the example verification, the application realizes the following technical effects:

[0417] 1. Data credibility enhancement: through on-chain writing, the calibrated error data, images and compensation values are non-tamperable and traceable;

[0418] 2. Error control ability improvement: in a complex fluctuation environment, the error control of the application is 200-300% better than that of the traditional method;

[0419] 3. Operation process transparency: the accuracy rate of abnormal operation identification is more than 85%, and the process supervision ability is significantly enhanced;

[0420] 4. Result automatic delivery: the report generation time is shortened from 30 minutes manually to 2 minutes automatically completed by the system;

[0421] 5. System robustness verification: in the case of environment temperature fluctuation ±2.5℃ and humidity ±8%RH, the system runs stably, and the compensation effect is obvious.

[0422] The above is the description of the embodiments of the application, through the above description of the disclosed embodiments, the person skilled in the art can realize or use the application. Various modifications of these embodiments will be apparent to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A camera-assisted based detection instrument field intelligent calibration method, characterized in that, Comprise the following steps in sequence: 1) Site and personnel legality confirmation: Obtain the calibration site coordinates through GPS and / or Beidou dual-mode positioning, and compare them with the pre-stored work order coordinates; and use face and voiceprint dual recognition to verify the identity of the calibration personnel uniquely; 2) Multi-camera panoramic acquisition and calibration: Arrange at least three high-resolution cameras with preset positions around the instrument being calibrated, obtain the internal and external parameters of the cameras through Zhang Zhengyou calibration method, and upload them to the cloud; start each camera synchronously to generate multi-view video stream with timestamp; 3) Visual-data fusion calibration: Call the convolutional neural network-OCR combined model to automatically recognize the standard instrument readings and the display values of the instrument being calibrated in the video stream, and write them into the local cache according to the one-to-one correspondence; calculate the first calibration error by differentiating the real-time recognition values and the standard calibration points; 3) Abnormal action and environment comprehensive discrimination: Use human pose estimation algorithm to detect whether the action sequence of the calibration personnel deviates from the pre-set working condition; simultaneously collect temperature, humidity, light and vibration data to compensate the first calibration error for environmental drift, and obtain the second calibration error; 4) Blockchain evidence storage and AR prompt: Calculate the image hash, environmental data and second calibration error generated in steps 2)-4) through SHA-3-256, and write them into the alliance chain, and digitally sign them by elliptic curve P-256; if the error or abnormal score exceeds the threshold, immediately display the correction instructions on the wearable AR terminal, and allow remote experts to mark them in real time on the AR screen through WebRTC; 5) Intelligent report generation: Based on the differential calculation results, the first frame screenshot of the camera evidence image and the environmental drift compensation model, automatically generate a structured calibration report with predictive maintenance suggestions, and add the on-chain transaction hash.

2. The method of claim 1, wherein, Step 2) specifically comprises the following steps: a) arrange at least three 4K industrial cameras along more than three non-collinear points around the instrument being calibrated, so that their optical axes intersect to form a minimum triangular network; b) use 9x6 checkerboard calibration targets to perform Zhang Zhengyou calibration on each camera in turn, obtain the internal parameter matrix K, distortion coefficients (k1, k2, p1, p2) and external parameter matrix [R|t], and calculate the SHA-3-256 hash of the parameter file and upload it to the cloud storage; c) synchronize the time bases of each camera system through IEEE-1588 precise clock, ensure that the difference between the triggering and acquisition times is not greater than 5ms, and write the nanosecond-level timestamp in the frame header; d) start recording video for each camera simultaneously under the same triggering instruction, output multi-view video stream with unified timestamp, and provide spatiotemporally consistent image data for subsequent visual-data fusion calibration.

3. The method of claim 1, wherein, Step 3) specifically includes the following steps: a) Collecting time-stamped I-frames at the same trigger time by at least three cameras synchronized by IEEE-1588 clock; b) Performing contrast-limited adaptive histogram equalization, bilateral filter denoising and YOLO-based ROI automatic cropping on the I-frames in turn; c) Sending the cropped images into the ResNet-BiLSTM-CTC convolution-OCR combined network to obtain the standard meter readings and the instrument readings to be calibrated and their confidence levels respectively; d) Writing the two types of readings into the local cache one by one according to the time stamp, and fusing the multi-camera results through a confidence-weighted voting mechanism; e) Calculating the difference between the fused readings and the preset standard calibration points to generate the first calibration error and archive the frame hash and confidence label. Step 4) specifically includes the following steps: a) Integrating the key I-frame hash, environmental baseline hash and second calibration error collected by the multi-camera into evidence metadata, calculating the SHA-3-256 digest and generating a digital signature using the ECDSA-P256 private key; b) Using the "consortium chain + IPFS" hierarchical strategy, packaging the metadata together with the IPFS content addresses of the video and sensor raw files into a transaction structure, writing it into the blockchain through BFT-DPoS consensus, and the block time is ≤5 seconds; c) When the compensation error or the comprehensive abnormal score is higher than the preset threshold of 0.1% FS or 0.4 points, a red correction instruction is immediately popped up on the wearable AR terminal, and remote experts are allowed to draw vector annotations in real time in the AR view through the WebRTC channel, and the on-site operator can continue the subsequent calibration process after completing the correction according to the annotations. The AR terminal automatically superimposes the real-time difference results of step C on the screen area of the instrument to be calibrated based on SLAM technology.

4. The method of claim 1, wherein, 6. The method of claim 1, characterized in that when the error exceeds the threshold value, the system automatically issues a freeze instruction to the enterprise ERP, locking the current calibration work order and preventing the instrument from being put into use; the structured calibration report contains a calibration credibility index, which is obtained by weighting the abnormal action score, environmental stability score and error residual score.

5. The method of claim 1, wherein, a) Positioning and identity verification unit, including GPS / Beidou module and face-voiceprint integrated identification terminal; b) Camera array unit, composed of at least three 4K cameras, with hardware synchronization and self-correction functions; c) Visual-data fusion engine, deployed on GPU edge gateway, used for performing OCR recognition, difference calculation and pose analysis; d) Environmental perception and compensation module, including temperature and humidity, light, vibration sensors and LSTM-based drift compensation submodule; e) Blockchain client module, used for writing image hash, compensation data and calibration error, and performing digital signature; f) AR collaboration module, supporting on-site visual guidance and real-time annotation by remote experts; g) Intelligent report generation and ERP interface module, used for outputting structured reports and interacting with enterprise information systems.

7. A field intelligent calibration system for a detection instrument implementing the method of any of claims 1-6, comprising: ​ ​ 8. The system of claim 11, wherein, The camera array unit supports H.265 encoding and RTSP push stream, and has low-illumination imaging capability in a scene of ≤10 lux; the vision-data fusion engine supports 60FPS real-time inference with a delay of ≤200ms; the environmental perception and compensation module has a temperature measurement accuracy of ±0.1°C and a relative humidity measurement accuracy of ±1%RH; the blockchain client module adopts the BFT-DPoS consensus algorithm, and the average block time is ≤5s; the AR collaboration module is compatible with the OpenXR standard and can run on mainstream AR glasses and smartphones. 9.A non-transitory computer readable storage medium having stored thereon a computer program, the computer program, when executed by a processor, causing the processor to implement the steps of the method of any one of claims 1-6. 10.A computer program product comprising program instructions that, when executed on an electronic device, cause the electronic device to perform the method of any one of claims 1-6.

Images