Intelligent acquisition instrument based on AI analysis and adaptive sampling and adaptive interrogation method

Abstract

The invention relates to the technical field of engineering monitoring, measurement and control, in particular to an intelligent acquisition instrument based on AI analysis and self-adaptive sampling and a self-adaptive interrogation method. According to the technical scheme, the device comprises a multi-source sensing interface module, an edge intelligent processing module, a self-adaptive sampling module, a communication module, a storage module, a key interrogation module, a power supply protection module and a structure packaging module, multi-dimensional feature fusion and edge decision making are achieved, the self-adaptive sampling module automatically adjusts sampling frequency according to an AI analysis result and an environment threshold value and has a dynamic threshold value adjusting function, the communication module supports various wireless protocols, and the key interrogation module can achieve automatic sensor recognition and instant data acquisition and uploading. The system can be widely applied to the fields of industrial monitoring, environment monitoring and equipment health management, and efficient and intelligent data acquisition and management are realized.

Description

Technical Field

[0001] This invention relates to the field of engineering monitoring and control technology, and in particular to an intelligent data acquisition instrument and adaptive sampling method based on AI analysis and adaptive sampling. Background Technology

[0002] In fields such as Industrial Internet of Things (IIoT), environmental monitoring, and equipment health management, data acquisition devices are key equipment for achieving state perception and data transmission. Traditional data acquisition devices have relatively simple functions, typically only responsible for forwarding the raw data collected by sensors to the cloud or host computer system without processing. This mode has many limitations: First, the continuous uploading of massive amounts of raw data puts enormous pressure on communication network bandwidth and cloud storage resources, and also leads to high power consumption; second, due to the lack of local computing power, the system cannot respond quickly to sudden abnormal events, and the fixed data acquisition frequency makes it difficult to optimize the sampling strategy between stable and abnormal states. This may result in missing critical high-frequency transient information, and may also lead to wasted energy and storage space when the data is stable.

[0003] Furthermore, the sensor access and identification process of existing data acquisition devices is cumbersome, typically requiring manual configuration of the address and parameters of each sensor. This leads to low deployment efficiency and a high risk of errors. In field operations, the lack of convenient local interaction mechanisms makes it difficult for users to proactively trigger real-time data acquisition or perform equipment status diagnostics. Additionally, traditional devices are prone to data loss during communication interruptions, and their anti-interference capabilities and physical protection levels in complex industrial environments are often insufficient, impacting system reliability and data integrity.

[0004] To address these issues, we propose an intelligent data acquisition instrument based on AI analysis and adaptive sampling, along with an adaptive recall method. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing an intelligent data acquisition instrument and an adaptive recall method based on AI analysis and adaptive sampling.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent data acquisition device based on AI analysis and adaptive sampling, comprising:

[0007] The multi-source sensing interface module includes various analog and digital signal interfaces for connecting various types of sensors. Specifically, it includes two 4-20mA analog signal input channels and two RS485 digital communication interfaces. The RS485 digital communication interfaces support connection methods to connect up to 32 sensors.

[0008] The edge intelligence processing module, connected to the multi-source sensing interface module, is configured to perform the following operations:

[0009] The interface is used to collect physical quantity data of the environment and equipment from the sensors; the built-in AI analysis model is run to perform real-time edge computing on the collected data, and the edge computing includes at least one of data filtering, anomaly detection, feature extraction, equipment fault prediction and early warning and operation mode recognition;

[0010] An adaptive sampling module is used to automatically adjust the data sampling frequency based on the output of the edge intelligent processing module and a preset environmental threshold.

[0011] The communication module supports multiple wireless communication protocols, including Wi-Fi, LoRa, 4G / 5G and NB-IoT, and is used to upload processed data to the cloud platform and receive remote commands.

[0012] The storage module is used to store program code, AI models and collected data, and automatically saves data when the network is interrupted. The data storage capacity is no less than 500,000 records.

[0013] The button recall module includes a built-in recall button for receiving manual trigger signals from the user;

[0014] The power supply protection module includes a built-in rechargeable battery and adjustable DC 5V, 12V, 18V, and 24V power output terminals, and is equipped with opto-isolation and a 1500W surge protection unit on the RS485 communication line.

[0015] The structural encapsulation module adopts an IP68 waterproof and dustproof fully enclosed design, and its operating temperature range is -40℃ to 85℃.

[0016] The edge intelligence processing module is further configured to automatically identify and register all connected sensors and their addresses in response to initial power-on and specific operations triggered by the recall button.

[0017] Preferably, the adaptive sampling module is specifically configured as follows:

[0018] When the monitored physical quantity data does not exceed the preset threshold, the system is in low power mode and samples at a first preset frequency; when the AI ​​analysis model detects data anomalies or the physical quantity data exceeds the preset threshold, the system automatically switches to high-frequency sampling mode and samples at a second preset frequency higher than the first preset frequency.

[0019] Furthermore, the adaptive sampling module also includes a dynamic threshold adjustment unit, which is used to correct the preset threshold in real time according to the rate of change of the environment, so as to avoid noise signals triggering high-frequency sampling.

[0020] Preferably, the edge intelligence processing module adopts an embedded AI model, which realizes feature learning and anomaly classification through a hybrid structure of convolutional neural network and support vector machine, for equipment health monitoring and fault early warning;

[0021] The edge intelligence processing module is also configured to perform data recovery functions, which retrieve locally stored data and upload it according to instructions received from the cloud platform within a specified time period.

[0022] The edge intelligent processing module also has a heterogeneous data synchronization function, which is used to unify the time calibration of temperature, humidity, acceleration, displacement, water level and vacuum signals to achieve multi-dimensional feature fusion.

[0023] Preferably, the communication module supports an edge-cloud collaborative computing architecture, and can adjust at least one of the local AI model parameters and update the sampling strategy according to cloud instructions.

[0024] Preferably, the key recall module is configured as follows:

[0025] When the user double-clicks the recall button, sensor address identification and initialization operations are performed.

[0026] When a user clicks the poll button, an instant data poll is performed and uploaded to the monitoring platform.

[0027] The time interval for double-clicking is a predetermined time, and the system indicates the recognition process by flashing and keeping the indicator light on the recall button, and the recognition is completed when the indicator light goes out.

[0028] The adaptive recall method for intelligent data acquisition instruments based on AI analysis and adaptive sampling includes the following steps:

[0029] Initialization process: When the device is powered on for the first time or when it is first assembled and used, the user triggers the call button twice in a row within a specified time. In response to this operation, the data acquisition instrument automatically scans, identifies and parses the address information of all the connected sensors.

[0030] Standard recall procedure: In subsequent use, when the user triggers the recall button once, the data acquisition device responds to this operation and performs a complete data acquisition and upload task;

[0031] Protection procedure: In the conventional recall procedure, the system enforces minimum time interval protection. After a single recall task is completed, a predetermined time interval must be elapsed before the next recall trigger is responded to, in order to prevent equipment damage caused by rapid continuous operation.

[0032] Edge intelligence analysis process: Feature extraction and anomaly identification are performed using embedded AI models;

[0033] Dynamic sampling control process: Automatically adjusts the sampling frequency based on edge intelligent analysis results to achieve adaptive switching between high and low frequencies;

[0034] Local caching and recall triggering process: Data is automatically cached when communication is abnormal, and users can trigger at least one of instant recall and emergency measurement by pressing a key;

[0035] Cloud-based collaborative management process: When the network is restored, the cached data is uploaded in time periods, and the AI ​​model and sampling parameters are adjusted based on cloud feedback.

[0036] Preferably, the AI ​​analysis model in the edge intelligent analysis process adopts a sliding window training and update mechanism, which can continuously optimize the prediction accuracy during equipment operation.

[0037] Preferably, the recall trigger includes two methods: button click and remote command. Button trigger prioritizes the execution of local recall tasks, while remote command prioritizes the execution of data synchronization tasks.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] This invention, by configuring multiple 4-20mA analog input channels and an RS485 digital communication interface, can flexibly connect up to 32 sensors, realizing unified acquisition and management of analog and digital signals, and adapting to the monitoring needs of multiple scenarios;

[0040] With its built-in embedded AI model (a hybrid structure of CNN and SVM), the data acquisition device can perform real-time analysis, filtering, feature extraction, fault prediction, and pattern recognition at the source of data generation. This greatly reduces the reliance on cloud computing, lowers network bandwidth usage and cloud service costs, and significantly improves the system's response speed to equipment anomalies and faults, providing core technical support for predictive maintenance.

[0041] The adaptive sampling module can dynamically switch sampling frequencies and intelligently adjust thresholds based on data status (whether it is abnormal or exceeds a threshold) and environmental change rate. It uses low-frequency sampling to reduce power consumption when data is stable and automatically switches to high-frequency sampling to capture details when anomalies are detected. This "on-demand sampling" strategy maximizes device battery life while ensuring no critical data is lost, making it particularly suitable for battery-powered field or mobile monitoring scenarios.

[0042] The innovative button-based recall module and automatic recognition mechanism make the deployment of sensor networks extremely simple. Users can complete the automatic identification and registration of all sensors with a simple double-click operation, avoiding complex manual configuration. A single click operation can trigger real-time data recall at any time, which greatly facilitates on-site debugging and emergency monitoring and improves human-computer interaction efficiency.

[0043] Multiple design features ensure stable system operation in harsh environments. The IP68-rated protective enclosure and wide operating temperature range enable it to adapt to harsh industrial and natural environments. The power supply protection module features multiple voltage outputs, opto-isolation, and surge protection units, which effectively improve the stability and lifespan of the system and connected sensors. The storage module automatically caches no less than 500,000 data entries when the network is interrupted and performs data retransmission after the network is restored, completely avoiding data loss.

[0044] The communication module supports multiple wireless protocols and has edge-cloud collaboration capabilities. The cloud can remotely issue commands to adjust local AI model parameters or update sampling strategies, enabling the entire system to continuously learn and optimize. It has the ability to remotely operate and maintain and iteratively upgrade. The heterogeneous data synchronization function aligns the data from multiple sources in time, laying a solid foundation for subsequent multi-dimensional data fusion and in-depth analysis.

[0045] The accompanying adaptive recall method combines AI edge analysis and dynamic sampling control to enable automatic response, data caching, and cloud-based collaborative management of devices under different operating conditions, exhibiting strong adaptability and fault tolerance. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the hardware architecture of the present invention;

[0047] Figure 2 This is a schematic diagram of the workflow of the method of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1

[0050] like Figure 1 As shown, the intelligent data acquisition device based on AI analysis and adaptive sampling proposed in this invention adopts a modular design concept. Each functional module exchanges data and transmits instructions through an internal bus, forming a complete edge intelligent sensing system, including:

[0051] Multi-source sensor interface module:

[0052] As the sensing front end of the system, this module adopts a highly integrated interface design: the 8-pin aviation plug interface adopts the industrial standard connector specification, and its pin definition has been carefully optimized: the red line (Pin1) is assigned as the +12V sensor power supply positive terminal, the orange line (Pin2) is the power supply ground line, and the yellow line (Pin3) and green line (Pin4) are respectively used as the A line and B line of RS485 communication.

[0053] The remaining blue, black, gray, and white lines (Pin 5-8) are reserved as expansion function pins, which can be flexibly configured according to the sensor type. In actual deployment, when connecting a vibrating wire sensor, the red / orange / blue / black four lines form a frequency signal acquisition loop, while the yellow / green / gray / white lines are used to read the temperature compensation signal; when connected to a smart pressure transmitter, the red / orange lines provide excitation power, and the yellow / green lines handle Modbus RTU protocol communication.

[0054] The 4-pin aviation connector is specifically designed for analog sensor input and uses a current loop transmission method to enhance anti-interference capability. The red and orange wires form the first 4-20mA signal channel, and the yellow and green wires form the second channel. Each input is equipped with a precision sampling resistor and a high-precision ADC (analog-to-digital converter) to achieve a measurement accuracy of 0.1%.

[0055] The two RS485 interfaces use transceiver chips that conform to the EIA / TIA-485 standard, support bus-type "daisy-chain" connection topology, and through a unique address automatic identification algorithm, a single bus can connect up to 32 smart sensor nodes with unique addresses.

[0056] Core Processing Unit

[0057] The core processing unit includes: an edge intelligent processing module and an adaptive sampling module, wherein:

[0058] Edge intelligent processing module

[0059] This module uses a 32-bit microcontroller with an ARM Cortex-M7 core as its core and is equipped with a specially optimized embedded AI inference framework. Its main functions are as follows:

[0060] Data Acquisition and Analysis: Upon system power-on or receipt of a specific initialization signal (double-clicking the recall button), the automatic enumeration process begins. The processor broadcasts a query command via the RS485 bus and employs an improved binary address scanning strategy to complete address identification and type registration for all online sensors within 3 seconds. During identification, the side button indicator light exhibits a breathing flash (frequency 2Hz), and after identification is complete, it remains constantly lit for 1 second before turning off, providing intuitive status feedback.

[0061] Low-level signal processing: A digital filter bank is implemented at the driver level, including but not limited to moving average filtering (for temperature signals), bandpass filtering (for vibration signals), and sliding window standard deviation calculation. In a preferred embodiment, anomaly detection can be performed using the following exemplary formula:

[0062]

[0063] in,

[0064] D i Indicates the current sampling point x i The standardized deviation coefficient,

[0065] μ t The mean value within the sliding time window.

[0066] σ t Standard deviation

[0067] When D i When the threshold value is greater than λ (where λ is a dynamic threshold, usually 2 to 3), the system determines that there is an abnormal trend at that moment and triggers a high-frequency sampling or alarm process.

[0068] Mid-level feature extraction: For vibration monitoring scenarios, time-domain features (root mean square value, peak value, kurtosis) and frequency-domain features (energy values ​​of specific frequency bands obtained through FFT transformation) are calculated in real time. In a preferred embodiment, the time-domain root mean square feature can be calculated by the following formula:

[0069]

[0070] in,

[0071] F RMS This represents the root mean square characteristic of the vibration signal;

[0072] N is the total number of sampling points.

[0073] x i Let be the i-th sampled value.

[0074] Simultaneously, to extract frequency domain energy features, the system performs a Fast Fourier Transform (FFT) on the signal to obtain the frequency domain power spectrum P(f). In an exemplary calculation:

[0075]

[0076] in,

[0077] E b Frequency band energy;

[0078] f1 and f2 are the upper and lower limits of the target frequency band.

[0079] By combining time-domain and frequency-domain features to form a feature vector F = [F RMS E b [, ...], are used as inputs to the AI ​​model to achieve intelligent recognition of device status.

[0080] High-level AI inference: A pruned and quantized convolutional neural network (CNN) model is ported to this system. This network accepts raw time-series data from multi-channel sensors as input and automatically learns and extracts deep features through its convolutional layers. These features are then fed into fully connected layers for device health status scoring and fault probability prediction. Simultaneously, the system computes time-domain and frequency-domain features (such as root mean square and frequency band energy) from the raw data in parallel and inputs them into a support vector machine (SVM) classifier for rapid identification of sudden anomalies. Finally, the outputs of the CNN and SVM are fused at the decision level to arrive at the final judgment. In an exemplary embodiment, the output feature mapping of the CNN part can be represented as:

[0081] z l =f(W l *x l-1 +b l )

[0082] in:

[0083] z l This is the output feature map of the l-th layer;

[0084] W l The kernel weight matrix;

[0085] * indicates a convolution operation;

[0086] f(·) is a nonlinear activation function (such as ReLU);

[0087] b l This is the bias vector.

[0088] After multiple convolutions and pooling, the extracted high-dimensional feature vector F is input into the SVM classifier. The decision function of the SVM can be expressed as:

[0089]

[0090] in:

[0091] K(F i F) is a kernel function (such as the radial basis function RBF);

[0092] α i b represents the model parameters obtained during the training phase;

[0093] y i For sample labels.

[0094] The features output by CNN describe the deep structural features of the device status, SVM performs fast boundary discrimination for sudden anomalies, and the fusion result of the two is used to generate the final health score or anomaly alarm signal.

[0095] Intelligent data management: Equipped with 128Mbytes of SPI Flash memory, it adopts a circular storage strategy, automatically overwriting the oldest data when the stored data reaches the 500,000-record limit. It features a complete data indexing mechanism, supporting fast retrieval by timestamp range, providing a foundation for data recovery functions.

[0096] Adaptive sampling module

[0097] This module achieves dynamic adjustment of the sampling frequency through close cooperation between software algorithms and hardware timers:

[0098] Dual-mode operation mechanism:

[0099] Basic monitoring mode: The system defaults to ultra-low power state and uses the RTC (real-time clock) wake-up mechanism to perform routine monitoring with a sampling granularity of 1 time / hour. In this mode, the main processor is in sleep state most of the time, only maintaining the operation of the real-time clock and some peripherals, and the power consumption of the whole machine is controlled below 300μA.

[0100] Intensive sampling mode: When the AI ​​analysis engine detects abnormal features or physical quantities exceeding dynamic thresholds, it immediately triggers an interrupt to wake up the entire system and initiates high-frequency sampling based on a hardware timer. The sampling frequency is software-adjustable within the range of 10Hz to 1kHz, and the duration is a configurable parameter (default 30 seconds), ensuring the capture of complete abnormal event waveforms;

[0101] Dynamic threshold optimization: The threshold adjustment unit adopts an adaptive learning mechanism, dynamically updating the trigger threshold based on the statistical characteristics (mean and standard deviation) of historical data and the gradient of environmental parameter changes. Specifically, for equipment in a stable operation phase, a relatively strict threshold (e.g., ±2σ) is set; when a gradual change in operating conditions is detected, the threshold range is appropriately widened (e.g., ±3σ) to effectively distinguish between real anomalies and normal fluctuations. In a preferred embodiment, the dynamic threshold can be calculated using the following exemplary formula:

[0102] T upper =μ t +k·σ t ,T lower =μ t -k·σ t

[0103] in:

[0104] μ t This represents the average of data over a recent period (sliding window).

[0105] σ t Standard deviation;

[0106] k is an adaptive coefficient, whose value is dynamically adjusted between 2 and 3 according to the environmental stability.

[0107] When real-time data x i Exceeding the range [T] lower T upper When this occurs, the system immediately switches to high-frequency sampling mode to capture potential abnormal signals.

[0108] Communication module

[0109] The communication module adopts a modular design, and the core board connects different communication modules through the MiniPCIe interface:

[0110] Wireless transmission solution: In areas with 4G / 5G network coverage, Quectel EC series modules are used to achieve high-speed data transmission; in outdoor scenarios where there is no mobile network but long-distance transmission is required, Semtech SX1276 chip is selected to achieve LoRa spread spectrum communication, with a transmission distance of up to 5 kilometers; for densely deployed indoor environments, Wi-Fi (ESP32 series) and Bluetooth 5.0 are supported for local data aggregation.

[0111] Edge-cloud collaboration mechanism: It is designed with a two-way command system, which can send configuration commands in JSON format from the cloud to dynamically adjust parameters such as confidence threshold and feature weight of edge AI model; at the same time, it supports differential upgrade technology, which transmits only the firmware difference part, significantly reducing the data traffic during OTA (over-the-air) upgrade.

[0112] Storage module

[0113] The storage module is the core of ensuring data reliability and intelligent system recovery. Its design goes beyond simple data recording functions, and it adopts an advanced hybrid storage architecture and intelligent management algorithms.

[0114] Hybrid storage architecture

[0115] High-speed cache: Employing 1MB of SRAM, this area temporarily stores raw sampled data, providing high-speed data access support for real-time computation and AI model inference in the edge intelligent processing module. Data in this area is released or transferred after processing is complete.

[0116] Large-capacity non-volatile storage area: The core storage unit uses a 128Mbit (16MB) SPI Flash memory. This storage area has undergone rigorous erase / write cycle testing to ensure write reliability throughout the device's lifespan. Based on 500,000 data records and an average of 256 bytes per record, the total storage capacity far exceeds design requirements, providing ample data buffer space for long-term deployment.

[0117] To ensure synchronized storage of multi-source data, in a preferred embodiment, the system uses global timestamp calibration to align signals with different sampling frequencies. The time calibration relationship can be exemplarily represented as follows:

[0118] t′ i =t i +Δt clk +ε i

[0119] in:

[0120] t i This is the original sampling time;

[0121] Δt clk This is the system clock offset (obtained via RTC correction or network time synchronization);

[0122] ε i This represents the sampling delay error between each sensor channel.

[0123] By periodically synchronizing and correcting, the time error ε of all channels is reduced. i The control is within milliseconds, thereby ensuring temporal consistency of heterogeneous data such as vibration, temperature, and pressure during fusion analysis.

[0124] Intelligent management algorithm

[0125] Circular storage and overwrite strategy: The firmware incorporates a highly efficient circular storage management algorithm. When the amount of stored data reaches a preset capacity limit (e.g., 90%), the system automatically initiates an overwrite mechanism, prioritizing the overwriting of the oldest non-critical data that has been successfully uploaded to the cloud. For critical data marked as "alarm" or "abnormal," the system places it in a protected storage area to prevent it from being overwritten by regular circular storage, ensuring the integrity of fault evidence.

[0126] Indexing and Fast Retrieval Mechanism: A metadata index containing a precise timestamp, sensor channel number, and data quality flags is generated for each stored data record. This index structure supports fast retrieval using a binary search method based on time range. When the network recovers and receives a "data recall" command from the cloud platform, the system can locate the target data block within milliseconds based on the time period parameter in the command, achieving efficient data recovery and uploading.

[0127] Furthermore, the storage module is equipped with a large-capacity tantalum capacitor as a backup power supply. When the system detects a main power failure or power outage, a hardware interrupt is immediately triggered, and the backup power supply can provide the storage system with at least 300ms of sustainment time. During this period, the processor urgently writes temporary data that has not yet been written to Flash from the cache, as well as the current file system state information, to non-volatile memory, completing an "emergency snapshot," thereby effectively preventing data loss or file system corruption caused by sudden power outages.

[0128] Key recall module

[0129] The physical buttons use waterproof tactile switches, and reliable detection is achieved through hardware debounce circuitry and software state machine.

[0130] Operation state machine: Defines three-state operation logic—single click (duration < 500ms), double click (interval between two presses 800ms-1500ms), and long press (duration > 3s). Each operation triggers a different response process and provides visual feedback through multi-color LED indicators;

[0131] Protective Timer: Implements a hardware watchdog mechanism at the software level. Once a rapid series of operations is detected (interval < 60 seconds), the system will ignore subsequent requests and display an error code on the indicator light (e.g., flashing 3 times rapidly) to prevent stress accumulation from damaging the sensor or interface circuit.

[0132] To prevent accidental triggering or stress damage caused by frequent operation, in a preferred embodiment, the validity of the button can be determined according to the following exemplary logical formula:

[0133]

[0134] in:

[0135] t press This represents the current button press duration.

[0136] t last This is the time since the last valid key press;

[0137] T min The minimum allowable interval time (e.g., 60 seconds);

[0138] V=1 indicates a valid operation, and V=0 indicates that the operation is ignored.

[0139] This logic effectively prevents continuous key presses within a short period of time. Combined with the LED error flashing indicator mechanism, it can significantly reduce the rate of accidental operation.

[0140] Power supply protection module

[0141] Power management architecture: Employs TIBQ series battery management chips for precise monitoring and protection of the built-in lithium thionyl chloride battery. The DC-DC conversion circuit provides four isolated power outputs, each with overcurrent and short-circuit protection.

[0142] Advanced protection circuitry: A three-level protection circuit consisting of a TVS (Transient Voltage Suppressor) array and a gas discharge tube is deployed at both ends of the RS485 bus, capable of withstanding a 1.5kW lightning surge as defined by the IEC61000-4-5 standard. Simultaneously, high-speed optocouplers achieve 2500Vrms of electrical isolation, effectively blocking ground loop interference.

[0143] Structural encapsulation module

[0144] The outer shell is made of PC+ABS engineering plastic and is fully sealed using ultrasonic welding. The connectors are J599 series military-grade aviation plugs, internally potted with special thermally conductive silicone, ensuring both IP68 protection and improved heat dissipation. Environmental testing has verified that the equipment starts normally at -40℃ and exhibits no performance degradation after 72 hours of continuous operation at 85℃.

[0145] Example 2

[0146] like Figure 2 As shown, the adaptive recall method for an intelligent data acquisition device based on AI analysis and adaptive sampling proposed in this invention, compared to Embodiment 1, further includes:

[0147] S1: Initialization process

[0148] After the device is initially installed and powered on, the operator presses the call button on the side twice within a one-second interval. The data acquisition unit responds to this operation by initiating an automatic scanning program, identifying all sensors connected via the RS485 bus, parsing their communication addresses, and completing system initialization. At this time, the button indicator light will display a specific pattern (such as flashing) to indicate that identification is in progress. Once the indicator light goes out, initialization is complete.

[0149] S2: Routine recall and protection procedures

[0150] In daily monitoring, if real-time data acquisition is required, the user can press the call button once. After the data acquisition device responds, it immediately collects data from all sensors, analyzes it intelligently at the edge, and uploads it to the cloud platform. To prevent damage to the device from frequent operations, the system software is forcibly configured to wait at least one minute after a call command is executed before responding to the next call request triggered by the button.

[0151] S3: Edge Intelligence Analysis Process

[0152] The collected data is first processed at the edge. An embedded AI model (CNN-SVM hybrid model) performs real-time analysis of the data, performing feature extraction and anomaly detection. This model uses a sliding window mechanism for incremental learning, which can continuously fine-tune the model parameters using newly collected data, improving the accuracy of predictive maintenance.

[0153] S4: Dynamic Sampling Control Process

[0154] The system automatically adjusts its sampling strategy based on the analysis results from S3. When no anomalies are detected, it maintains low-power sampling (1 time / hour); once an anomaly is detected or data exceeds limits, it immediately switches to high-frequency sampling mode. Simultaneously, the dynamic threshold adjustment unit adaptively adjusts the threshold for triggering high-frequency sampling based on environmental trends.

[0155] S5: Local caching and recall triggering process

[0156] When the communication module detects a network interruption, all processed data is automatically saved to local storage. At this time, users can still perform local data retrieval by clicking a button for emergency situation assessment. Retrieval can be triggered by local button presses and remote commands, with the system assigning higher real-time response priority to local button presses.

[0157] S6: Cloud-based collaborative management process

[0158] Once the network is restored, the system will automatically, or upon receiving instructions from the cloud platform, upload the cached, timestamped data from the outage in batches. The cloud platform can then perform in-depth analysis on the aggregated data and distribute optimized AI model parameters or new sampling strategies to the edge, completing a closed loop of edge-cloud collaborative optimization.

[0159] The above specific embodiments are merely several preferred embodiments of the present invention. Based on the technical solutions of the present invention and the relevant teachings of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

[0160] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. An intelligent data acquisition instrument based on AI analysis and adaptive sampling, characterized in that, include: The multi-source sensing interface module includes various analog and digital signal interfaces for connecting various types of sensors. Specifically, it includes two 4-20mA analog signal input channels and two RS485 digital communication interfaces. The RS485 digital communication interfaces support connection methods to connect up to 32 sensors. The edge intelligence processing module, connected to the multi-source sensing interface module, is configured to perform the following operations: The interface is used to collect physical quantity data of the environment and equipment from the sensors; the built-in AI analysis model is run to perform real-time edge computing on the collected data, and the edge computing includes at least one of data filtering, anomaly detection, feature extraction, equipment fault prediction and early warning and operation mode recognition; An adaptive sampling module is used to automatically adjust the data sampling frequency based on the output of the edge intelligent processing module and a preset environmental threshold. The communication module supports multiple wireless communication protocols, including Wi-Fi, LoRa, 4G / 5G and NB-IoT, and is used to upload processed data to the cloud platform and receive remote commands. The storage module is used to store program code, AI models and collected data, and automatically saves data when the network is interrupted. The data storage capacity is no less than 500,000 records. The button recall module includes a built-in recall button for receiving manual trigger signals from the user; The power supply protection module includes a built-in rechargeable battery and adjustable DC 5V, 12V, 18V, and 24V power output terminals, and is equipped with opto-isolation and a 1500W surge protection unit on the RS485 communication line. The structural encapsulation module adopts an IP68 waterproof and dustproof fully enclosed design, and its operating temperature range is -40℃ to 85℃. The edge intelligence processing module is further configured to automatically identify and register all connected sensors and their addresses in response to initial power-on and specific operations triggered by the recall button.

2. The intelligent data acquisition device based on AI analysis and adaptive sampling according to claim 1, characterized in that: The adaptive sampling module is specifically configured as follows: When the monitored physical quantity data does not exceed the preset threshold, the system is in low power mode and samples at a first preset frequency; when the AI ​​analysis model detects data anomalies or the physical quantity data exceeds the preset threshold, the system automatically switches to high-frequency sampling mode and samples at a second preset frequency higher than the first preset frequency. Furthermore, the adaptive sampling module also includes a dynamic threshold adjustment unit, which is used to correct the preset threshold in real time according to the rate of change of the environment, so as to avoid noise signals triggering high-frequency sampling.

3. The intelligent data acquisition device based on AI analysis and adaptive sampling according to claim 1, characterized in that: The edge intelligent processing module adopts an embedded AI model, which uses a hybrid structure of convolutional neural network and support vector machine to achieve feature learning and anomaly classification for equipment health monitoring and fault early warning. The edge intelligent processing module is also configured to perform data recovery function, which retrieves locally stored data and uploads it according to instructions received from the cloud platform in a specified time period. The edge intelligent processing module also has a heterogeneous data synchronization function, which is used to unify the time calibration of temperature, humidity, acceleration, displacement, water level and vacuum signals to achieve multi-dimensional feature fusion.

4. The intelligent data acquisition device based on AI analysis and adaptive sampling according to claim 1, characterized in that: The communication module supports an edge-cloud collaborative computing architecture and can adjust at least one of the local AI model parameters and update the sampling strategy according to cloud instructions.

5. The intelligent data acquisition device based on AI analysis and adaptive sampling according to claim 1, characterized in that: The key recall module is configured as follows: When the user double-clicks the recall button, sensor address identification and initialization operations are performed. When a user clicks the poll button, an instant data poll is performed and uploaded to the monitoring platform. The time interval for double-clicking is a predetermined time, and the system indicates the recognition process by flashing and keeping the indicator light on the recall button, and the recognition is completed when the indicator light goes out.

6. An adaptive recall method for an intelligent data acquisition instrument based on AI analysis and adaptive sampling, characterized in that, Includes the following steps: Initialization process: When the device is powered on for the first time or when it is first assembled and used, the user triggers the call button twice in a row within a specified time. In response to this operation, the data acquisition instrument automatically scans, identifies and parses the address information of all the connected sensors. Standard recall procedure: In subsequent use, when the user triggers the recall button once, the data acquisition device responds to this operation and performs a complete data acquisition and upload task; Protection procedure: In the conventional recall procedure, the system enforces minimum time interval protection. After a single recall task is completed, a predetermined time interval must be elapsed before the next recall trigger is responded to, in order to prevent equipment damage caused by rapid continuous operation. Edge intelligence analysis process: Feature extraction and anomaly identification are performed using embedded AI models; Dynamic sampling control process: Automatically adjusts the sampling frequency based on edge intelligent analysis results to achieve adaptive switching between high and low frequencies; Local caching and recall triggering process: Data is automatically cached when communication is abnormal, and users can trigger at least one of instant recall and emergency measurement by pressing a key; Cloud-based collaborative management process: When the network is restored, the cached data is uploaded in time periods, and the AI ​​model and sampling parameters are adjusted based on cloud feedback.

7. The adaptive recall method of the intelligent data acquisition instrument based on AI analysis and adaptive sampling according to claim 6, characterized in that: The AI ​​analysis model in the edge intelligent analysis process adopts a sliding window training and update mechanism, which can continuously optimize the prediction accuracy during equipment operation.

8. The adaptive recall method of the intelligent data acquisition instrument based on AI analysis and adaptive sampling according to claim 6, characterized in that: The recall trigger includes two methods: button click and remote command. Button trigger prioritizes the execution of local recall tasks, while remote command prioritizes the execution of data synchronization tasks.

Images