A method for constructing an electronic identity card of a coal-to-oil machine pump

Abstract

A method for constructing electronic ID cards for coal-to-oil pumps, belonging to the field of intelligent operation and maintenance technology for coal chemical equipment, generates a hierarchical unique code including process segment code, hazard zone level, solids content calibration segment, and hash check code; employs a double-layer stainless steel vacuum-insulated RFID tag integrating a K-type thermocouple and a MEMS accelerometer, which can withstand 300℃ and the scouring of solids-containing oil slurry; constructs a digital twin archive containing a static library, a dynamic library, and a solids content-corrected Weibull wear model; the inspection terminal has a built-in 1D-CNN lightweight model to locally predict wear probability and only uploads abnormal data; the background implements four-level early warning based on the remaining lifespan probability; combines three-code fusion and blockchain for spare parts traceability; and achieves full lifecycle management through multi-physics field digital twin joint simulation and model iteration. This invention improves the survivability of electronic ID cards and the level of intelligent operation and maintenance under extreme operating conditions.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology of coal chemical equipment, specifically relating to a full life cycle intelligent system for coal-to-oil pumps. Background Technology

[0002] Coal-to-oil technology uses coal as raw material to produce oil products and petrochemical products through chemical processing. It includes two technical routes: direct coal liquefaction and indirect coal liquefaction. In the coal-to-oil production process, pump equipment (such as reactor circulation pumps, hydrogenation feed pumps, and high-temperature slurry pumps) is one of the most critical moving parts. Taking the direct coal liquefaction process as an example, the reactor circulation pump needs to operate under high temperature of 455℃ and high pressure of 20MPa, conveying an oil slurry containing up to 50% coal powder particles. Furthermore, these pumps typically operate individually with no backup; a failure would lead to an unplanned shutdown of the entire plant, resulting in significant economic losses.

[0003] Currently, coal-to-oil enterprises generally manage pumps and machinery using methods such as manual registration, paper ledgers, or QR code labels. These methods have the following prominent problems: 1. Electronic identification tags are difficult to survive long-term under extreme operating conditions: Ordinary RFID tags and QR code tags cannot withstand the high temperatures (>250℃), high-solids-content oil slurry scouring, strong corrosion, and strong vibration environments of coal-to-oil pumps, and usually fail within weeks to months. Although some existing technologies have high-temperature resistant RFID tags (such as those using ceramic encapsulation), they do not have self-sensing capabilities for temperature and vibration, and cannot provide real-time sensing data for health management.

[0004] 2. Fragmented equipment management information and lack of a closed-loop system covering the entire equipment lifecycle: Existing management methods mostly remain at the level of registering basic equipment information. Operational data, maintenance records, and spare parts replacement information are scattered across different systems, making it impossible to form a dynamic electronic file with a unique identifier at its core. Some enterprises have introduced asset management systems, but manual data entry is still required, data is lagging, and it is impossible to achieve automatic traceability of the entire process from procurement to disposal.

[0005] 3. Lack of wear prediction and early warning capabilities for coal-to-oil operation: The remaining life of coal-to-oil pumps is affected by multiple factors such as the solid content of the medium, temperature, pressure, and vibration. Most of the early warnings of existing general equipment management systems are based on fixed cycles or single parameter thresholds (such as monitoring only the vibration amplitude), and cannot dynamically adjust the prediction model according to the solid content. The false alarm rate and the missed alarm rate are high, making it difficult to achieve predictive maintenance.

[0006] Therefore, there is an urgent need to invent a full-lifecycle intelligent system for coal-to-oil pumps. Through technologies such as high-temperature and corrosion-resistant self-sensing tags, solids-content adaptive wear models, edge intelligent prediction, and digital twin iteration, the system can achieve digital and intelligent management of the entire lifecycle of coal-to-oil pumps from procurement to disposal, ensuring the safe and stable operation of the equipment. Summary of the Invention

[0007] To overcome the problems existing in the background art, the present invention provides a method for constructing an electronic ID card for a coal-to-oil pump.

[0008] To achieve the above objectives, the present invention provides a method for constructing an electronic ID card for a coal-to-oil pump, comprising the following steps: Step A: Generate an adaptive unique code for the operating conditions of the coal-to-oil machine pump. The process section code, hazard zone level, medium solids content calibration section, and equipment tag number of the coal-to-oil pump are obtained, and a timestamp-based anti-counterfeiting verification code is generated based on the SM3 hash algorithm. The above fields are then concatenated in a fixed order to form a hierarchical unique code. The solids content calibration section is used to identify the solid particle concentration range (5%~65%) of the medium transported by the pump.

[0009] Step B: Install integrated self-sensing high-temperature and corrosion-resistant RFID electronic tags The RFID electronic tag is encapsulated in a double-layer stainless steel metal shell and a silicon nitride ceramic core. A vacuum heat insulation layer is set between the shell and the core, and the inner wall of the shell is coated with a nano-alumina heat insulation coating. A K-type thin-film thermocouple temperature sensing unit and a MEMS triaxial accelerometer vibration sensing unit are integrated on the ceramic core. All three are encapsulated in the same tag body, and the overall thickness of the tag is ≤8mm. Tag operating temperature range: -40℃~300℃ (continuous), can withstand 350℃ (30 minutes); protection rating IP69K, explosion-proof mark ExibIICT4; The tag chip storage area includes an EPC area (496-bit), a TID area (256-bit), and a user data area (2K-bit). The user data area is further divided into a static information sub-area, a dynamic fluctuation sub-area, and a wear characteristic sub-area.

[0010] Step C: Construct a digital twin archive database for pumps and motors Using the unique code generated in step A as the primary index, establish the following in the backend server: Static information database: stores basic device attributes, technical parameters, drawings and documents; Dynamic fluctuation library: Stores real-time temperature, vibration time-domain / frequency-domain characteristics, and solids content data of the medium; Wear feature library: based on the Weibull distribution model, and with the introduction of a solids content correction factor K. s =1+0.02×(C s -20%), C sThe real-time solids content ranges from 5% to 65%. The remaining life probability distribution function of the core components of the computer pump is used, and historical wear curves are stored. The core components are: impeller, bearing, and seal. Step D: Double Label Fixing and Parameter Binding The label is attached to the bearing housing of the pump or the non-high temperature area of ​​the pump body using high-temperature resistant epoxy resin structural adhesive, and then fixed in place by mechanical anti-loosening buckles. The unique code and static parameters generated in step A are written into the user data area of ​​the tag using an explosion-proof handheld terminal, and the tag ID is bound to the corresponding model in the digital twin archive database described in step C.

[0011] Step E: Local Inspection and Predictive Synchronization Based on Lightweight Convolutional Neural Networks Inspection personnel used an explosion-proof handheld terminal with a built-in 1D-CNN lightweight convolutional neural network model to read real-time temperature and vibration data from the tags via near-field communication. The terminal performs the following calculations locally: The vibration signal is truncated into a 3.2-second window, input into a 1D-CNN model, and the output is the wear anomaly probability P. w (0~1); If P w If the value is ≥0.7 or the temperature change rate exceeds 5℃ / min, the terminal will automatically transmit the key data frames (vibration spectrum, P...). w Values ​​and temperature curves are uploaded to the backend server to update the digital twin database; otherwise, they are stored locally and not uploaded. Step F: Dynamic four-level early warning and remaining life classification Based on the remaining lifespan probability distribution in the wear feature database, the backend server automatically classifies the warning status into four levels and sends them to the relevant personnel's terminals: Green: Remaining lifespan > 90% Yellow: 50% < Remaining lifespan ≤ 90% Orange: 10% < Remaining lifespan ≤ 50% Red: Remaining lifespan ≤ 10% The calculation of the remaining lifetime prediction value also takes into account the real-time solids content C. s Vibration energy weighting value E v Temperature cumulative effect T acc The warning threshold is dynamically adjusted according to the operating conditions.

[0012] Step G: Three-code integrated blockchain spare parts traceability Each spare part is given a 3-code fusion tag consisting of an RFID built-in code, a QR code, and a serial number, and the spare part information (production batch, warehousing time, performance parameters) is stored on the blockchain. When spare parts are installed on the pump, the replacement record and the corresponding blockchain evidence hash value are written into the user data area of ​​the pump's electronic tag and the backend database by scanning the spare part tag and the pump tag, forming an immutable two-way traceability chain between the spare parts and the pump.

[0013] Step H: Multiphysics Digital Twin Co-simulation and Model Iteration Using the digital twin archive database established in step C, three physical field sub-models are constructed: a fluid dynamics model, an impeller erosion and wear model, and a bearing fatigue life model. Using a unique code as an index, real-time dynamic fluctuation data is used as a boundary condition input to the sub-model for joint simulation, and a comprehensive health index (0~100) is output. After each maintenance, the shape and scale parameters of the Weibull distribution are corrected in reverse based on the measured wear and replacement parts records. At the same time, transfer learning is used to update the 1D-CNN model to achieve iterative evolution of the model.

[0014] Preferably, the vacuum degree of the vacuum insulation layer is ≤10⁻²Pa, and the wall thickness of the double stainless steel outer shell is 0.5~1.0mm, and the thickness of the intermediate vacuum layer is 0.3~0.8mm.

[0015] Preferably, the thickness of the K-type thin-film thermocouple is ≤0.3mm, its hot junction is integrated on the surface of the ceramic core, and the cold junction compensation adopts the PN junction temperature sensor built into the label, with a temperature measurement accuracy of ±1.5℃.

[0016] Preferably, the MEMS triaxial accelerometer has a sampling rate ≥25.6kHz and a range of ±50g, and can extract the pump impeller pass frequency (BPF) and its sidebands for identifying erosion wear characteristics.

[0017] Preferably, the 1D-CNN lightweight convolutional neural network model includes 3 convolutional layers, 2 pooling layers and 2 fully connected layers, with ReLU as the activation function and Sigmoid as the output layer. The model is deployed on the MCU (ARM Cortex-M7) of the explosion-proof handheld terminal through TensorFlow Lite for Microcontrollers, with a single inference time of <0.5 seconds and power consumption of <100mW.

[0018] Preferably, the solids content correction factor K s The upper limit of the correction is 2.0 (when C). s When ≥65%, the lower limit is 0.8 (when C s (When ≤5%), the remaining lifetime probability distribution function adopts a two-parameter Weibull distribution, with its scale parameter η and K... s Inversely proportional: η = η0 / K s , where η0 is the dimensional parameter under rated operating conditions.

[0019] Preferably, the formula for calculating the comprehensive health index HI is: Among them, P w R represents the probability of abnormal wear. l V represents the percentage of remaining lifespan. c Indicates the degree of compliance with vibration intensity, T m This indicates the temperature margin; the weighting coefficients mentioned above (0.4, 0.3, 0.2, 0.1) can be optimized and adjusted based on the pump type through historical data learning.

[0020] Preferably, the explosion-proof handheld terminal also has a built-in vibration spectrum real-time analysis module, which can perform FFT transformation on vibration data locally, extract the energy ratio of the first harmonic, second harmonic and sideband, and upload abnormal spectrum segments as key data frames, with the uploaded data volume compressed by more than 90% compared with the original waveform.

[0021] A system for constructing an electronic ID card for a coal-to-oil generator pump includes: RFID electronic tag: as described in step B of claim 1; Explosion-proof handheld inspection terminal: Built-in 1D-CNN lightweight model and spectrum analysis module, used to perform local inspection and predictive synchronization as described in step E of claim 1; Background digital twin server: used to perform the database construction, early warning and model iteration described in steps C, F and H of claim 1; Blockchain-based evidence storage platform: used to perform spare parts traceability and evidence storage as described in step G of claim 1. The above-mentioned units are connected via encrypted wireless network or near-field communication to jointly achieve electronic identity management and health prediction for the entire lifecycle of the coal-to-oil pump.

[0022] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention employs a composite encapsulation structure consisting of a double-layer stainless steel outer shell, a vacuum insulation layer, a nano-alumina insulation coating, and a silicon nitride ceramic core. This extends the operating temperature range of the RFID electronic tag to -40℃ to 300℃ (continuously), and it can withstand short-term impacts of 350℃ (30 minutes), achieving an IP69K protection rating and an explosion-proof mark of ExibIICT4. Actual testing shows that this tag can operate stably for ≥3 years in the high-temperature, solid-containing oil slurry environment of coal-to-oil production, overcoming the technical prejudice that "RFID / electronic tags cannot be used in high-temperature, solid-containing conditions in coal-to-oil production," and filling the technological gap in this field for high-temperature resistant self-sensing electronic ID cards.

[0023] 2. This invention integrates a K-type thin-film thermocouple temperature sensing unit and a MEMS triaxial accelerometer vibration sensing unit into the same package of an RFID electronic tag, with a thickness ≤8mm, solving the engineering challenge of miniaturized integration of multiple sensors under high-temperature conditions. The matching explosion-proof handheld terminal has a built-in 1D-CNN lightweight convolutional neural network model (model ≤500KB), which can perform real-time inference of wear anomaly probability locally, with a single inference time of <0.5 seconds and power consumption of <100mW. It only uploads key data frames, saving more than 95% of communication traffic compared to the traditional full-upload solution, and supports offline inspection, significantly improving the intelligence level and network adaptability of explosion-proof area operations.

[0024] 3. This invention utilizes the unique solids content (C) of the medium in coal-to-oil conversion. s By introducing the solids content as an independent variable into the Weibull wear model and establishing a solids content correction factor Ks = 1 + 0.02 × (Cs - 20%), the remaining life prediction error is reduced from ±35% in conventional methods to within ±15%. This mathematical modeling method is not publicly available in existing technologies and can dynamically adjust the life assessment benchmark according to actual working conditions, providing reliable data support for predictive maintenance.

[0025] 4. This invention constructs three physical field sub-models: a fluid dynamics model, an impeller erosion and wear model, and a bearing fatigue life model. These are then jointly simulated using real-time dynamic fluctuation data to output a comprehensive health index (HI). After each maintenance, the system reverse-corrects the Weibull distribution parameters and CNN model weights based on the measured wear amount, forming a closed-loop digital twin system of "perception → diagnosis → prediction → iteration." The model's prediction accuracy continuously improves with usage time, achieving self-optimization for the entire equipment lifecycle management.

[0026] Instruction manual illustrations Figure 1 A flowchart illustrating the overall process for constructing electronic ID cards for coal-to-oil generator pumps; Figure 2 This is a schematic diagram of a hierarchical unique coding structure; Figure 3 A schematic diagram of the architecture of a digital twin archive database; Figure 4 Flowchart for local inspection and edge prediction; Figure 5 A dynamic four-level early warning and decision-making flowchart; Figure 6 A flowchart for the three-code integrated blockchain spare parts traceability process; Figure 7 Multiphysics digital twin co-simulation architecture diagram. Detailed Implementation

[0027] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto. Example

[0028] Please refer to Figures 1-7 Taking a large-scale coal direct liquefaction to oil project as an example, an electronic identification system was established for the circulating pump of the hydrogenation reactor in the coal liquefaction unit, and full life-cycle health management was implemented. The process medium of this circulating pump is a high-temperature solid-containing oil slurry with a temperature of 455℃, an operating pressure of 20MPa, and a coal powder particle concentration of 45%~55%. The electronic identification tag is installed on the side of the pump body bearing housing (about 300mm from the high-temperature medium area, with a measured surface temperature of 80~120℃ at the installation point), and does not directly contact the 455℃ medium.

[0029] Step 1: Generate an adaptive unique code for the operating conditions of the coal-to-oil machine pump. This circulating pump belongs to the hydrogenation section of the coal liquefaction unit. The hazard level is ExⅡCT4. The calibration range for the solid content of the medium is 50% (the actual operating fluctuation range is 45%~55%, and the median value is taken as the calibration range). The equipment tag number is RCP-2101.

[0030] The system generates a multi-segment, hierarchical unique code through a background coding service. The format is: <Process Segment Code>—<Hazard Zone Level>—<Solids Content Calibration Segment>—<Equipment Tag Number>—<Timestamp>—<Check Code>, totaling 6 segments, each separated by a hyphen. The process segment code and equipment tag number indicate the equipment's hierarchical level, the hazard zone level indicates the safety level, the solids content calibration segment indicates the operating condition level, and the timestamp and check code ensure uniqueness and anti-counterfeiting features.

[0031] The specific generation process is as follows: Process section code: ML (coal liquefaction) Hazard zone level code: T4 Solid content calibration range: 50 Equipment tag number: RCP2101 Timestamp: 20260424120000 (accurate to the second) Anti-counterfeiting verification code: After concatenating the above fields, the first 8 bits are taken as the verification code using the SM3 hash algorithm. The final unique code generated is: ML-T4-50-RCP2101-20260424120000-A3F2C8D1 The code generation process employs a database auto-incrementing sequence combined with a snowflake algorithm to ensure no duplication across multiple concurrent devices. If a hash collision occurs during generation, the system automatically adds a timestamp in microseconds and recalculates until a globally unique code is generated.

[0032] Step 2: Install integrated self-sensing high-temperature and corrosion-resistant RFID electronic tags Customized RFID electronic tags are used, with the following structure: Outer shell: Double-layer stainless steel (SUS316L), outer layer wall thickness 0.8mm, inner layer wall thickness 0.6mm, intermediate vacuum insulation layer thickness 0.5mm, vacuum degree ≤10⁻²Pa. The inner wall of the outer shell is coated with a nano-alumina heat insulation coating with a thickness of 20μm. This composite structure can effectively block the installation point (80~120℃) and the radiant heat of the medium (455℃), ensuring that the operating environment temperature of the chip inside the tag is below 85℃. As a further durability improvement solution, the outer shell material can also be selected from high-temperature nickel-based alloys (such as Inconel 625), the coating thickness can be increased to 50μm, and the vacuum insulation layer can be filled with aerogel powder (thermal conductivity <0.02W / (m·K)).

[0033] Core: Silicon nitride ceramic substrate, dimensions 10mm×8mm×1.5mm.

[0034] Temperature sensing unit: K-type thin film thermocouple, 0.25mm thick, with the hot junction integrated on the surface of the ceramic core. Cold junction compensation is achieved by using a PN junction temperature sensor built into the label. The temperature measurement accuracy is ±1.5℃ (normal temperature conditions), and after high-temperature calibration, the accuracy is ±2.0℃ at 300℃.

[0035] Vibration sensing unit: MEMS triaxial accelerometer (model: ADXL355), sampling rate 25.6kHz, range ±50g, zero bias stability ±10mg, sensitivity 2.4mg / LSB.

[0036] Chip: ImpinjM730, EPC area 496-bit, TID area 256-bit, user data area 2K-bit.

[0037] Temperature resistance: The label body has a continuous operating temperature range of -40℃ to 300℃ and can withstand short-term impact at 350℃. After accelerated aging testing at 350℃ for 30 minutes, the performance is evaluated as follows: RF reading distance: Initially 1.5m, after testing 1.4m, change rate <7%; Temperature sensing unit: The temperature measurement accuracy has been improved from ±1.5℃ to ±1.8℃, but it is still within the allowable range of ±2.0℃ in engineering. Vibration sensing unit: Zero bias stability changed from ±10mg to ±15mg, sensitivity change <3%, sampling rate remained unchanged at 25.6kHz, and vibration monitoring function was normal.

[0038] Protection rating: IP69K, explosion-proof mark: ExibIICT4.

[0039] Label dimensions: 28mm in diameter, 7.5mm in thickness, and 18g in weight.

[0040] Step 3: Build a digital twin archive database for pumps and motors Create three database instances on the backend server: (1) Static database (MySQL) Store the following information about this circulating pump: Basic attributes: Equipment name, model, manufacturer, serial number, purchase date, and commissioning date; Technical parameters: Rated flow rate 1100m³ / h, rated head 220m, rated power 2500kW, rated speed 2980r / min; Supporting information: Motor model (YB3-450-4), mechanical seal model (H75-110), bearing model (SKF6324 / C3); Drawings and documents: equipment assembly drawings, piping system diagrams, operation manuals, etc. (saved in OSS object storage mode, with the URL and MD5 checksum stored in the database).

[0041] (2) Dynamic fluctuation library (InfluxDB time series database) The following data is written in real time: Vibration values: Effective values ​​of X / Y / Z triaxial acceleration (mm / s²); Temperature: bearing temperature, pump body temperature (°C); Solid content of the medium: Real-time data from online particle size analyzer (%), ranging from 5% to 65%; Process parameters: inlet pressure, outlet pressure, flow rate, motor current.

[0042] The data sampling interval adopts an adaptive variable strategy: under normal operating conditions (wear anomaly probability P) w <0.5) Maintain a 1-minute interval; when P is detected w When the value is ≥0.5 or the temperature change rate is >5℃ / min, automatically switch to 10-second encrypted sampling until P. w If it falls below 0.5 and remains below that level for 30 minutes, then resume the 1-minute interval.

[0043] (3) Wear feature library (MongoDB document database) The system stores the remaining life probability distribution calculated based on the Weibull distribution model, as well as historical wear curves. Models are established for key components of the circulating pump (impeller, bearings, mechanical seal).

[0044] Taking the impeller as an example: Under rated operating conditions, the Weibull dimensional parameter η0 = 32000 hours and the shape parameter β = 2.5.

[0045] Real-time solids content C s =48%, calculate correction factor Ks =1+0.02×(C s -20%).

[0046] Here, 20% is the design baseline value for the solids content of the oil slurry in the direct coal liquefaction process (based on the process package specifications); when the solids content is lower than 20%, the correction factor is less than 1 (set lower limit 0.8); when the solids content is higher than 20%, the correction factor is greater than 1 (upper limit 2.0). The coefficient 0.02 can be determined through laboratory erosion tests or field data regression.

[0047] Actual scale parameter η = η0 / K s =32000 / 1.56≈20513 hours.

[0048] Remaining lifetime probability distribution function: The current running time is t=8500 hours, and the median remaining lifetime is approximately 20513×ln(2)^(1 / 2.5)-8500≈7680 hours.

[0049] The database uses AES-256 encryption for storage, with daily full backups and real-time incremental backups, and backup files are stored off-site. Alternatively, the national standard SM4 encryption algorithm can be used.

[0050] Step 4: Double Label Fixing and Parameter Binding Perform the following operations on the side of the circulating pump bearing housing (approximately 300mm from the high-temperature medium area, with a measured surface temperature of 80~120℃ at the installation point): 1. Clean the surface to remove oil and oxide layers.

[0051] 2. Apply high-temperature resistant epoxy structural adhesive (choose a two-component epoxy adhesive that has been tested to withstand temperatures above 200°C for a long period and 250°C for a short period, such as Henkel Loctite, 3M, or similar domestic products), and affix the label to the designated position.

[0052] 3. Use M4 stainless steel mechanical anti-loosening locks to tighten the mounting ears on the edge of the label to the pre-drilled threaded holes in the pump body, and apply thread locking agent after tightening.

[0053] 4. Using an industrial-grade handheld terminal that conforms to the ExibIICT4Gb explosion-proof standard (e.g., an explosion-proof PDA running Android system and integrating NFC read / write function), the NFC function is used to conduct near-field communication with RFID electronic tags.

[0054] 5. Bind the tag ID to the corresponding model in the digital twin archive database in the backend system.

[0055] Step 5: Local Inspection and Predictive Synchronization Based on Lightweight Convolutional Neural Networks (1) Training of 1D-CNN model Training data sources: Historical operating data of similar circulating pumps from the target factory over the past 5 years were collected, including vibration waveforms (sampling rate 25.6kHz), temperature curves, solids content records, and corresponding actual maintenance records (normal / minor wear / severe wear / failure), serving as the basic training sample set. To improve the model's generalization ability and cover a wider range of operating conditions, the training data can be further expanded, including but not limited to: data from similar equipment at other production bases of the same enterprise, publicly available failure case data from the same industry, simulation data generated through mechanism simulation models, and extreme operating condition samples synthesized using generative adversarial networks (GANs) (such as solids content > 60%, temperature surge > 10℃ / min, etc.). Through the above methods, a diverse training set covering the complete failure mode space can be constructed.

[0056] Sample size description: In practical applications, the total sample size can range from several thousand to hundreds of thousands of samples, depending on the amount of available data. In this embodiment, the basic sample size is approximately 32,000 samples, which can be expanded to over 100,000 samples to meet the model training requirements.

[0057] Data preprocessing: The vibration signal is truncated in a 3.2-second window, the DC component is removed (using mean subtraction), and the signal is normalized to the [0,1] interval.

[0058] Model architecture: Input layer: 81920×1 Conv1D(32,kernel_size=64,stride=8,activation='relu') MaxPooling1D(pool_size=4) Conv1D(64,kernel_size=32,stride=4,activation='relu') MaxPooling1D(pool_size=4) Conv1D(128,kernel_size=16,stride=2,activation='relu') GlobalAveragePooling1D() Dense(64,activation='relu') Dense(1,activation='sigmoid') → Outputs the probability of wear anomalies, P. w Training method: TensorFlow 2.0 was used with the Adam optimizer, a learning rate of 0.001, a batch size of 64, and 100 iterations. The training set, validation set, and test set were split in a 7:1.5:1.5 ratio. The test set accuracy was 92.3%, and the AUC was 0.96.

[0059] Model compression: The model is converted into a C++ byte array using TensorFlowLite for Microcontrollers. The model size is approximately 480KB, which meets the ≤500KB requirement.

[0060] (2) Inspection process Inspection personnel conduct inspections every 8 hours. Specific procedures are as follows: When the handheld explosion-proof terminal is brought close to the tag (distance <10cm), the terminal uses 13.56MHz NFC to automatically read the real-time temperature and vibration data inside the tag (the most recent 1 minute continuous waveform; if it is in encrypted sampling state, the most recent 10 seconds waveform is read).

[0061] The terminal runs a 1D-CNN model locally and outputs P. w Values. Actual measured data: Vibration effective value on X-axis 4.2 mm / s, Y-axis 5.1 mm / s, Z-axis 3.8 mm / s, bearing temperature 89℃, P w =0.65.

[0062] Because P w =0.65<0.7, and the temperature change rate<5℃ / min, which is considered a normal fluctuation. The terminal does not upload data, but only stores logs locally.

[0063] If a certain inspection P w =0.82 (exceeding the 0.7 threshold), the terminal automatically sends the key data frames (vibration spectrum FFT results, P...) w The data (values ​​and temperature curves) are compressed and uploaded to the backend server. The amount of data uploaded is about 12KB per upload, which is 96% compressed compared to the original waveform (about 320KB).

[0064] The terminal uses an ARM Cortex-M7 processor (400MHz), with a measured single inference time of 0.42 seconds, power consumption of 72mW, and battery life sufficient for 8 hours of continuous operation.

[0065] Step 6: Dynamic four-level early warning and remaining life classification The backend server updates the remaining lifespan probability distribution every 4 hours based on the wear feature database data and performs early warning judgment.

[0066] Calculation conditions: Median life expectancy forecast: 7680 hours It has been in operation for 8500 hours, with a rated lifespan of 32000 hours. Remaining lifespan percentage = 7680 / 32000 = 24%. (1) Vibration energy weighting value E v Calculation: To comprehensively evaluate the impact of vibration intensity on wear life of the circulating pump, this invention introduces a vibration energy weighting value E. v Calculate using the following formula: E v =w x ·V x² +w y ·V y² +w z ·V z² in: V x V y V z These are the effective values ​​of vibration velocity in the horizontal radial (X-direction), vertical radial (Y-direction), and axial (Z-direction) directions, measured on the pump body bearing housing, in mm / s. The vibration signals in these three directions reflect the characteristics of different fault modes, such as rotor imbalance, misalignment, bearing failure, and fluid disturbance.

[0067] V x² V y² V z² These represent the squares of the three effective values ​​of the above three velocities, with units of (mm / s)², and are used to characterize vibration energy. wx, wy, and wz are the corresponding weighting coefficients, which are dimensionless and reflect the degree of contribution of vibration in each direction to the overall health of the pump.

[0068] Methods for determining weighting coefficients: The weighting coefficients reflect the contribution of vibrations in different directions to the overall health of the pump. They can be determined using various methods based on the pump type, structural characteristics, and historical fault data, including but not limited to: Expert experience-based assignment method: This method involves equipment engineers directly setting values ​​based on their field experience. For example, for centrifugal pumps, axial vibration is usually the most significant indicator of thrust bearing wear, and the value can be assigned to the bearing. z Set to a larger value.

[0069] Data-driven approach: Collect historical vibration data and wear failure records of similar pumps and motors, and use principal component analysis (PCA) or partial correlation analysis to calculate the correlation between vibration amplitude in each direction and actual wear amount. Use the normalized correlation coefficient as the weighting coefficient.

[0070] Based on the Analytic Hierarchy Process (AHP): Multiple equipment experts were invited to compare the importance of vibrations in different directions pairwise, construct a judgment matrix, calculate the eigenvectors, and then normalize them to obtain the weight coefficients.

[0071] Adaptive dynamic adjustment method: Based on the actual operating data and maintenance records of the pumps, the system uses algorithms such as gradient descent or Bayesian optimization to update the weight coefficients periodically (e.g., quarterly) to better reflect the actual wear characteristics of the current equipment.

[0072] In this embodiment, based on the mechanism analysis of the vibration characteristics of the centrifugal pump, axial vibration (Z-axis) directly reflects the axial movement of the impeller and the wear state of the thrust bearing, and has the most significant impact on the safe operation of the circulating pump; horizontal radial vibration (X-axis) mainly reflects rotor imbalance and foundation loosening; vertical radial vibration (Y-axis) is more affected by fluid pulsation and has relatively lower sensitivity to faults. Therefore, a weighting coefficient of w is chosen. z =0.6, w x =0.3, w y =0.1. In practical applications, the above weighting coefficients can be adaptively adjusted according to the equipment type and operating conditions.

[0073] Example calculation: Assume the effective value of the vibration velocity measured during a certain inspection is: V x =2.8mm / s, V y =2.1mm / s, V z =1.6mm / s, substitute into the above weighting coefficients for calculation: E v =0.3×(2.8)²+0.1×(2.1)²+0.6×(1.6)²=0.3×7.84+0.1×4.41+0.6×2.56=2.352+0.441+1.536= 4.329(mm / s)² The higher this value, the higher the current vibration energy, and the greater the negative impact on the remaining lifespan of the pump. The backend system will then... v As one of the inputs, it participates in the calculation of the comprehensive health index HI and the dynamic adjustment of the early warning threshold.

[0074] (2) Temperature cumulative effect T acc Calculation To assess the cumulative effects of long-term over-temperature operation of bearings, the temperature cumulative effect T is defined. acc : in: T represents the real-time temperature (°C) measured by the K-type thin-film thermocouple from step 2. T base=80℃ (reference temperature, based on the upper limit of the recommended operating temperature of the grease). ∫(TT base) dt is the integral of the temperature exceeding the baseline value within the statistical time window Δt (unit: °C·h); Δt is taken from the past 7 days (168 hours), with a sampling interval of 1 minute, and is calculated using numerical integration.

[0075] Example calculation: The bearing temperature fluctuated between 85 and 92°C over the past 7 days, exceeding Tbase by an average of about 8°C. After integration, Tacc = 2.3°C (equivalent average overtemperature). The larger Tacc is, the more severe the overtemperature, which accelerates grease aging and metal fatigue.

[0076] In this embodiment, T acc Primarily used for auxiliary analysis and the temperature margin T of the comprehensive health index HI m This is a reference and is not directly involved in the remaining lifetime quantification calculation. In subsequent model iterations, T can be established based on actual data. acc Empirical relationship with lifespan correction.

[0077] (3) Probability distribution of remaining lifetime and early warning level In this embodiment, the remaining lifespan of the impeller is modeled using a two-parameter Weibull distribution: The scale parameter η is introduced by adding a solids content correction factor K based on the rated operating condition reference value η0. s get: in: η0 = 32000 hours (design life); C s The real-time solids content (%) is given by C0 = 20% (design baseline value) and α = 0.02 (empirical coefficient, which can be determined through laboratory erosion tests or field data regression). β=2.5 (shape parameter, based on statistical data of similar pump failures).

[0078] Calculation conditions: The current running time is t=8500 hours, C s =48%, K s =1+0.02×(48-20)=1.56,η=32000 / 1.56≈20513 hours; Median remaining lifespan T remaining ≈20513×ln(2)^(1 / 2.5)-8500≈7680 hours; Design total life T total=32000 hours, remaining lifespan percentage R l =7680 / 32000×100%=24%; E v =4.33(mm / s)², T acc =2.3℃.

[0079] Warning level classification: According to R l =24%, the current state is determined to be an orange alert. The system automatically pushes the alert information to the workshop director and equipment engineer, and generates a pre-maintenance work order. The alert record (time, level, handler, and result) is stored in the database for subsequent model iterations.

[0080] Step 7: Three-code integrated blockchain spare parts traceability (1) Generation of unique spare parts identifiers Every critical spare part entering the warehouse (such as mechanical seals, impellers, and bearings) must be generated with a three-code integrated identifier: RFID built-in code: A globally unique UDI (Unique Device Identification) is assigned by the spare parts management system. It is 24 characters long and contains spare parts type code, batch number, and serial number.

[0081] QR code: Adopting the QR Code standard, the encoded content includes the traceability platform URL (including UDI parameters), the first 8 digits of the anti-counterfeiting hash value, and the spare parts manufacturer code.

[0082] Serial number: The serial number is directly marked on the metal surface of the spare part body using laser etching technology (the position does not affect installation and operation), and the font depth is ≥0.2mm to ensure resistance to oil stains and washing.

[0083] The three codes are interlocked using the SM3 hash algorithm: the hash value is calculated by concatenating the RFID built-in code and the serial number, and the first 8 bits of the hash value are used as the verification field of the QR code; by scanning any one code, the system can query the complete information corresponding to the other two codes.

[0084] (2) Spare parts information blockchain storage When spare parts are received into the warehouse, the warehouse manager uses an explosion-proof handheld terminal to scan three codes, and the system automatically submits the hash values ​​of the following information to the blockchain evidence storage platform: Basic information for spare parts: manufacturer, date of manufacture, batch number, performance parameters (such as spring pressure and end face roughness of mechanical seals); Certificates of conformity and quality inspection reports: stored as PDF files on IPFS (InterPlanetary File System), with their CID (Content Identifier) ​​uploaded to the blockchain along with them; Inbound information: Inbound time, storage location number, and operator's number.

[0085] The blockchain platform adopts a consortium blockchain architecture, with participating nodes including: procurement department, warehousing department, maintenance department, and corporate oversight and auditing department. The consensus mechanism uses Raft (suitable for enterprise-level internal applications). After successful notarization, the platform returns the transaction hash (TxHash) and block height, which the system then writes to the spare parts archive database.

[0086] (3) Spare parts installation and associated binding When equipment maintenance requires replacement of spare parts, maintenance personnel shall perform the following operations: Authentication: Log in to the system using a fingerprint or work card via an explosion-proof handheld terminal.

[0087] Scan to associate: Scan the equipment label (to read the unique code of the pump) and the three codes of the spare part to be installed in sequence (choose one, and the system will automatically associate the other two codes).

[0088] Authenticity verification: The system queries the blockchain evidence storage platform based on the URI in the QR code and compares the spare parts information stored on the chain with the current scan result to see if they are consistent; at the same time, it checks the status code on the chain - if the spare parts status is "installed" or "scrapped", the installation operation is rejected and an alarm is triggered.

[0089] Record generation: Maintenance personnel fill in the reason for replacement (selectable from the drop-down menu, such as "wear exceeds limit", "seal leakage" or "preventive replacement"). The system automatically generates a replacement record, which includes: pump code, UDI of new and old spare parts, replacement time, operator number, and vibration characteristic values ​​before / after replacement (extracted from the local inspection record in step 5).

[0090] On-chain confirmation: The hash value of the replacement record is uploaded to the blockchain evidence storage platform. To ensure successful upload, the terminal employs a retry mechanism: after sending a request, it waits for the platform to return a TxHash; if no return is received within 10 seconds, it automatically retryes up to 3 times. After receiving the TxHash, the terminal queries the platform for the confirmation status of the transaction (requiring confirmation from at least 2 nodes). Upon successful confirmation, the TxHash is written to the user data area (wear characteristic sub-area) of the pump's electronic tag, and the spare part status is updated to "Installed - Pump Code". All operation logs are stored locally on the terminal and uploaded to the audit server weekly.

[0091] (4) Traceability query Any authorized person (including equipment engineers, maintenance managers, and auditors) can check the complete lifecycle of spare parts through the following methods: Scan device tags: View a list of all currently installed spare parts and their replacement history.

[0092] Scan the spare part's QR code to view its factory information, warehousing record, which pump it was installed on, reason for replacement, operator, and corresponding blockchain-based notarized TxHash.

[0093] The traceability page also displays the on-chain storage status of spare parts (such as "Storage valid, block height 1234567"), which makes it easier for auditors to verify the authenticity and integrity of the data.

[0094] (5) Exception handling If the spare parts information is found to be inconsistent with the on-chain evidence (such as hash value mismatch or abnormal status code), the system will automatically trigger the following process: A "suspected counterfeit spare parts" work order is generated on the management platform, the batch number of the spare parts is marked, and the order is pushed to the procurement department and the quality inspection department.

[0095] This spare part is prohibited from being used for installation; already installed parts are marked "Pending Review".

[0096] Record abnormal events to the blockchain (only record the event type and UDI, without involving trade secrets).

[0097] Step 8: Multiphysics Digital Twin Co-simulation and Model Iteration (1) Architecture of digital twin model This embodiment constructs a digital twin system containing three physical field sub-models, as follows: Fluid Dynamics (CFD) Model: The Navier-Stokes equations for the flow field within the pump are solved using the finite volume method. Input boundary conditions include: inlet pressure, outlet pressure, medium density (varying with solids content), medium viscosity (varying with temperature and solids content), solid particle size distribution, and inlet concentration. Model outputs: velocity field within the pump, pressure distribution, solid particle trajectories, wall impact velocities, and impact angles.

[0098] Impeller erosion wear model: Based on the particle impact parameters output by CFD, the Finnie erosion model is used to calculate the wear rate at various locations on the impeller surface. The calculation formula is as follows: Where ER is the wear rate (mm / year), V p Where α is the particle impact velocity (m / s), α is the impact angle (°), and C is the particle impact velocity (m / s). s Here, α represents the solids content (%), k and n are material constants (calibrated by erosion tests; for steel, n is typically taken as 2.3~2.5), and f(α) is an angle function, classically expressed as f(α) = sin(2α) / 2 (the maximum value occurs at approximately 45°). Model output: a cloud map showing the wear depth distribution on the impeller surface, as well as the maximum wear depth and its location.

[0099] Bearing fatigue life model: The corrected reference life L10m of the bearing is calculated according to ISO281:2007 standard, taking into account correction factors such as reliability coefficient, material coefficient, lubrication condition coefficient, and contamination coefficient. Input parameters include: bearing geometric parameters (pitch circle diameter, rolling element diameter, contact angle), measured vibration acceleration spectrum, lubricating oil viscosity, and load spectrum (estimated based on pump outlet pressure fluctuations). Output: Probability distribution of the remaining life of the bearing (using Weibull distribution, with shape parameter β set to 1.5).

[0100] (2) Multiphysics coupling method The three sub-models exchange data using a loosely coupled approach: Each CFD model outputs the solid phase impact parameter distribution for each run (approximately 2 hours of computation time), which serves as input for the erosion model.

[0101] The erosion model is updated (once per hour) to calculate the impeller wear depth. When the maximum wear depth exceeds 5% of the initial thickness, the worn impeller geometry (surface coordinates after mesh deformation) is fed back to the CFD model to recalculate the flow field, so as to reflect the change of the flow field caused by wear.

[0102] The bearing model runs independently and is updated every 24 hours. Its prediction results (remaining life percentage) are used as one of the inputs to the overall health HI.

[0103] (3) Calculation of the Comprehensive Health Index (HI) The Comprehensive Health Index (HI) comprehensively reflects the overall health status of the pump and motor, and is calculated using the following formula: Where: P w Wear anomaly probability, output of the 1D-CNN model in step 5; T remaining The current predicted median remaining lifespan is calculated using the Weibull model in step 6. T total Total design life of equipment; static information pre-stored in database. Vc, vibration intensity compliance, is the normalized ratio of the measured effective value of vibration to the limit specified in ISO 10816-3; Tm, temperature margin, (alarm threshold - measured temperature) / alarm threshold; The weighting coefficients, 0.4, 0.3, 0.2, and 0.1, are determined based on the Analytic Hierarchy Process (AHP) and can be adaptively adjusted according to the equipment type.

[0104] Calculation example in this embodiment: P w =0.65 (from step 5) T remaining=7680 hours, T total =32000 hours, R l =0.24 The measured effective vibration value was 4.7 mm / s. ISO 10816-3 specifies an upper limit of 4.5 mm / s for zone C and a lower limit of 7.1 mm / s for zone D. c =1-(4.7-4.5) / (7.1-4.5)≈0.92 The measured temperature of the bearing was 89℃, and the alarm threshold was 95℃. m =(95-89) / 95≈0.063 Substitute into the formula: HI=0.4×(1-0.65)+0.3×0.24+0.2×0.92+0.1×0.063=0.4×0.35+0.072+0.184+0.0063=0.14+0.072+0.184+0.0063= 0.402 The HI value ranges from 0 to 1, with values ​​closer to 1 indicating better health. A result of 0.402 indicates the equipment is in a "sub-healthy" state and requires maintenance.

[0105] (4) Model iteration and self-learning After each maintenance is completed, the system iteratively updates the model based on the actual wear data in the maintenance records: Triggering conditions: The maintenance work order is closed, and the maintenance personnel have filled in the actual measured wear amount (such as the reduction value of impeller thickness, the increase value of bearing clearance) and the information of the replaced parts.

[0106] Update content: Weibull scale parameter η update: Using the actual operational lifespan (hours from commissioning to replacement) as the new observation, the scale and shape parameters are refitted using maximum likelihood estimation. If sufficient samples are lacking, a Bayesian update method combined with the prior distribution is used for correction.

[0107] Update of solids content correction factor coefficient α: based on multiple groups (C s The actual wear acceleration data points were used to refit the α value (original α=0.02) using linear regression, making the corrected formula closer to the actual field situation.

[0108] 1D-CNN model transfer learning: Vibration data (including normal and abnormal states) from a period before and after the maintenance node are used as new training samples. An incremental learning method is adopted (the learning rate is reduced to 1 / 10 of the original value, and only the last two fully connected layers are fine-tuned) to update the model weights. The updated model is pushed to all explosion-proof handheld terminals via OTA (over-the-air) and automatically synchronized during the next inspection.

[0109] HI weighting coefficient self-optimization: Collecting data from several maintenance records (P... w ,R l V c ,T m The formula includes a health status grading system (excellent / good / average / poor) and a manual assessment system. The gradient descent method is used to optimize the four weight coefficients in the HI formula to maximize the consistency between HI and actual health status.

[0110] Iteration frequency: Model iteration is performed immediately after major repairs (replacing impellers or bearings); it is performed in batches every quarter after routine maintenance (such as lubrication and tightening). A model version number is generated after each iteration for easy traceability.

[0111] (5) Visualization and application of digital twins The backend digital twin system provides a 3D visualization interface to display the following content: The pump body 3D model is rendered in real time, and the impeller wear thermal map is shown (the red area represents the high wear area).

[0112] Time-domain waveforms and spectrum diagrams of vibration in each direction, with characteristic frequencies marked (such as blade passage frequency and bearing failure frequency).

[0113] The remaining lifetime prediction curve (including the 95% confidence interval) shows a dynamic extension over operating time.

[0114] Historical warning records and maintenance recommendations list.

[0115] This visual interface supports access from both PC and mobile devices, enabling equipment administrators to remotely monitor and analyze data.

[0116] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A method for constructing an electronic ID card for a coal-to-oil pump, characterized in that, Includes the following steps: Step A: Generate an adaptive unique code for the operating conditions of the coal-to-oil machine pump. The process section code, hazard level, medium solids content calibration section, and equipment tag number of the coal-to-oil pump are obtained, and a timestamp-based anti-counterfeiting verification code is generated based on the SM3 hash algorithm. The above fields are then concatenated in a fixed order to form a hierarchical unique code. The solids content calibration section is used to identify the solid particle concentration range (5%~65%) of the medium transported by the pump. Step B: Install integrated self-sensing high-temperature and corrosion-resistant RFID electronic tags The RFID electronic tag is encapsulated in a double-layer stainless steel metal shell and a silicon nitride ceramic core. A vacuum heat insulation layer is set between the shell and the core, and the inner wall of the shell is coated with a nano-alumina heat insulation coating. A K-type thin-film thermocouple temperature sensing unit and a MEMS triaxial accelerometer vibration sensing unit are integrated on the ceramic core. All three are encapsulated in the same tag body, and the overall thickness of the tag is ≤8mm. Tag operating temperature range: -40℃~300℃ (continuous), can withstand 350℃ (30 minutes); protection rating IP69K, explosion-proof mark ExibIICT4; The tag chip storage area includes an EPC area (496-bit), a TID area (256-bit), and a user data area (2K-bit). The user data area is further divided into a static information sub-area, a dynamic fluctuation sub-area, and a wear characteristic sub-area. Step C: Construct a digital twin archive database for pumps and motors Using the unique code generated in step A as the primary index, establish the following in the backend server: Static information database: stores basic device attributes, technical parameters, drawings and documents; Dynamic fluctuation library: Stores real-time temperature, vibration time-domain / frequency-domain characteristics, and solids content data of the medium; Wear feature library: based on the Weibull distribution model, and with the introduction of a solids content correction factor K. s =1+0.02×(C s -20%), C s The real-time solids content ranges from 5% to 65%. The remaining life probability distribution function of the core components of the computer pump is used, and historical wear curves are stored. The core components are: impeller, bearing, and seal. Step D: Double Label Fixing and Parameter Binding The label is attached to the bearing housing of the pump or the non-high temperature area of ​​the pump body using high-temperature resistant epoxy resin structural adhesive, and then fixed in place by mechanical anti-loosening buckles. The unique code and static parameters generated in step A are written into the user data area of ​​the tag using an explosion-proof handheld terminal. At the same time, the tag ID is bound to the corresponding model in the digital twin archive database described in step C. Step E: Local Inspection and Predictive Synchronization Based on Lightweight Convolutional Neural Networks Inspection personnel used an explosion-proof handheld terminal with a built-in 1D-CNN lightweight convolutional neural network model to read real-time temperature and vibration data from the tags via near-field communication. The terminal performs the following calculations locally: The vibration signal is truncated into a 3.2-second window, input into a 1D-CNN model, and the output is the wear anomaly probability P. w (0~1); If P w If the value is ≥0.7 or the temperature change rate exceeds 5℃ / min, the terminal will automatically transmit the key data frames (vibration spectrum, P...). w Values ​​and temperature curves are uploaded to the backend server to update the digital twin database; otherwise, they are stored locally and not uploaded. Step F: Dynamic four-level early warning and remaining life classification Based on the remaining lifespan probability distribution in the wear feature database, the backend server automatically classifies the warning status into four levels and sends them to the relevant personnel's terminals: Green: Remaining lifespan > 90% Yellow: 50% < Remaining lifespan ≤ 90% Orange: 10% < Remaining lifespan ≤ 50% Red: Remaining lifespan ≤ 10% The calculation of the remaining lifetime prediction value also takes into account the real-time solids content C. s Vibration energy weighting value E v Temperature cumulative effect T acc The warning threshold is dynamically adjusted according to the operating conditions; Step G: Three-code integrated blockchain spare parts traceability Each spare part is given a 3-code fusion tag consisting of an RFID built-in code, a QR code, and a serial number, and the spare part information (production batch, warehousing time, performance parameters) is stored on the blockchain. When spare parts are installed on the pump, the replacement record and the corresponding blockchain evidence hash value are written into the user data area of ​​the pump electronic tag and the backend database by scanning the spare part tag and the pump tag, forming an immutable two-way traceability chain of "spare parts-pump". Step H: Multiphysics Digital Twin Co-simulation and Model Iteration Using the digital twin archive database established in step C, three physical field sub-models are constructed: a fluid dynamics model, an impeller erosion and wear model, and a bearing fatigue life model. Using a unique code as an index, real-time dynamic fluctuation data is used as a boundary condition input to the sub-model for joint simulation, and a comprehensive health index (0~100) is output. After each maintenance, the shape and scale parameters of the Weibull distribution are corrected in reverse based on the measured wear and replacement parts records. At the same time, transfer learning is used to update the 1D-CNN model to achieve iterative evolution of the model.

2. The method for constructing an electronic ID card for a coal-to-oil pump according to claim 1, characterized in that, The vacuum degree of the vacuum insulation layer is ≤10⁻²Pa, and the wall thickness of the double stainless steel shell is 0.5~1.0mm, while the thickness of the intermediate vacuum layer is 0.3~0.8mm.

3. The method for constructing an electronic ID card for a coal-to-oil pump according to claim 1, characterized in that, The K-type thin-film thermocouple has a thickness of ≤0.3mm, and its hot junction is integrated into the surface of the ceramic core. Cold junction compensation is achieved by using a PN junction temperature sensor built into the label, with a temperature measurement accuracy of ±1.5℃.

4. The method for constructing an electronic ID card for a coal-to-oil pump according to claim 1, characterized in that, The MEMS triaxial accelerometer has a sampling rate of ≥25.6kHz and a range of ±50g. It can extract the pump impeller pass frequency (BPF) and its sidebands for identifying erosion wear characteristics.

5. The method according to claim 1, characterized in that, The 1D-CNN lightweight convolutional neural network model contains 3 convolutional layers, 2 pooling layers and 2 fully connected layers, with ReLU activation function and Sigmoid function for output layer. The model is deployed on the MCU (ARM Cortex-M7) of the explosion-proof handheld terminal through TensorFlow Lite for Microcontrollers, with a single inference time of <0.5 seconds and power consumption of <100mW.

6. The method for constructing an electronic ID card for a coal-to-oil pump according to claim 1, characterized in that, The solids content correction factor K s The upper limit of the correction is 2.0 (when C). s When ≥65%, the lower limit is 0.8 (when C s (When ≤5%), the remaining lifetime probability distribution function adopts a two-parameter Weibull distribution, with its scale parameter η and K... s Inversely proportional: η = η0 / K s , where η0 is the dimensional parameter under rated operating conditions.

7. The method according to claim 1, characterized in that, The formula for calculating the comprehensive health index HI is as follows: Among them, P w R represents the probability of abnormal wear. l V represents the percentage of remaining lifespan. c Indicates the degree of compliance with vibration intensity, T m This indicates the temperature margin; the weighting coefficients mentioned above (0.4, 0.3, 0.2, 0.1) can be optimized and adjusted through self-learning based on historical data according to the pump type.

8. The method for constructing an electronic ID card for a coal-to-oil pump according to claim 1, characterized in that, The explosion-proof handheld terminal also has a built-in vibration spectrum real-time analysis module, which can perform FFT transformation on vibration data locally, extract the energy ratio of the first harmonic, second harmonic and sideband, and upload abnormal spectrum segments as key data frames. The amount of uploaded data is compressed by more than 90% compared with the original waveform.

9. A system for constructing an electronic ID card for a coal-to-oil pump, characterized in that, include: RFID electronic tag: as described in step B of claim 1; Explosion-proof handheld inspection terminal: Built-in 1D-CNN lightweight model and spectrum analysis module, used to perform local inspection and predictive synchronization as described in step E of claim 1; Background digital twin server: used to perform the database construction, early warning and model iteration described in steps C, F and H of claim 1; Blockchain-based evidence storage platform: used to perform spare parts traceability and evidence storage as described in step G of claim 1; The aforementioned units are connected via encrypted wireless networks or near-field communication to jointly achieve electronic ID management and health prediction for the entire life cycle of coal-to-oil pumps.

Images