An internet of things-based intelligent monitoring system and method for frost heaving force in cold region geotechnical engineering

Abstract

The present application relates to geotechnical engineering test equipment and engineering monitoring technical field, specifically to a kind of intelligent monitoring system and method of frost heaving force of cold region geotechnical engineering based on Internet of Things.The system includes distributed monitoring node and remote cloud platform;The monitoring node includes: sensing module, signal conversion module, main control module, wireless communication module, power supply module and local display module;The system provided by the present application realizes remote, real-time, automatic monitoring to rock mass crack frost heaving force, and can predict the development trend of crack, and timely issue early warning when data is abnormal, greatly improve the initiative and foresight of engineering safety management.The method provided by the present application is accurate, can realize real-time processing, trend prediction and intelligent early warning to frost heaving force data, and has broad application prospect in the field of cold region geotechnical engineering.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering testing equipment and engineering monitoring technology, specifically to an intelligent monitoring system and method for frost heave force in cold-region geotechnical engineering based on the Internet of Things. Background Technology

[0002] In cold-region geotechnical engineering, especially in high-altitude frozen rock areas, freeze-thaw cycles continuously degrade the mechanical properties of rock masses, becoming a core contributing factor to engineering disasters such as slope instability, tunnel lining cracking, and roadbed frost heave and thaw settlement. To investigate the mechanical behavior of rock masses and the evolution of frost heave force under freeze-thaw conditions, conventional techniques in the industry include: indoor testing using high-low temperature alternating test chambers to simulate natural freeze-thaw processes, monitoring changes in frost heave force in rock samples using pressure and displacement sensors, and conducting mechanical property tests using rock mechanics testing machines; and in field engineering, deploying monitoring sensors at rock mass cracks, collecting data on-site, and manually retrieving and analyzing it to monitor frost heave force. Long-term, accurate, and stable testing of frost heave force is a core prerequisite for evaluating the frost heave deformation and stress evolution of rock masses and for achieving early warning of engineering disasters.

[0003] Currently, pressure sensors and crack gauges are commonly used in indoor experiments and field engineering to monitor stress and deformation changes in rock samples / rock masses during freeze-thaw processes, thereby calculating frost heave force. However, existing monitoring systems can only complete basic data acquisition and transmission, lacking specialized preprocessing algorithms for frost heave force signals. They lack intelligent identification of freeze-thaw states and short-term trend prediction capabilities for frost heave force, and can only achieve passive data recording. They cannot provide proactive early warning for engineering safety management and are ill-suited for the long-term unattended monitoring needs of projects in uninhabited areas and remote cold regions. Furthermore, traditional data preprocessing uses a single temperature compensation coefficient and fixed filtering methods, which cannot adapt to the full temperature range of cold regions, including extremely low temperatures, negative temperatures, and positive temperatures. This easily leads to misjudging legitimate frost heave / thaw contraction signal mutations caused by water-ice phase change during freeze-thaw processes as outliers, resulting in the loss of effective data and distorted monitoring results, failing to provide reliable data support for engineering analysis.

[0004] Therefore, in order to meet the construction and operation and maintenance needs of frozen rock engineering in cold regions, there is an urgent need to develop a dedicated monitoring system for frost heave force that is adapted to the special environment of cold regions, is reliable in installation, accurate in measurement, and integrates specialized signal processing and intelligent analysis algorithms, along with complete monitoring methods, to provide core technical support for the quantitative analysis of rock mass frost heave mechanical behavior and the safety management of cold region engineering. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an intelligent monitoring system and method for frost heave force of cold-region geotechnical engineering with long-term monitoring and safety early warning functions for frost heave force of engineering such as high-altitude cold-region rock slopes, tunnels, and roadbeds, and which can be used in conjunction with indoor rock freeze-thaw mechanics tests.

[0006] To solve the above problems, the technical solution adopted by the present invention is as follows:

[0007] A smart monitoring system for frost heave force in cold-region geotechnical engineering based on the Internet of Things, the key features of which include: distributed monitoring nodes and a remote cloud platform;

[0008] The monitoring node includes: a sensing module, a signal conversion module, a main control module, a wireless communication module, a power supply module, and a local display module;

[0009] The sensing module includes a frost heave force sensing module and a temperature sensing module. The frost heave force sensing module is fixed to both sides of the crack in the rock mass to be monitored or the indoor rock sample, and is used to convert the width change or stress change at the crack into an electrical signal. The temperature sensing module collects the ambient temperature information and converts the ambient temperature information into an electrical signal.

[0010] The signal conversion module amplifies, filters, and performs analog-to-digital conversion on the electrical signal, and outputs a digital signal to the main control module.

[0011] The main control module has a built-in frost heave analysis algorithm to preprocess the collected data, identify the freeze-thaw state, predict trends, and assess risks, and generate early warning decisions; the wireless communication module uploads the processed data to the cloud platform and receives control commands from the cloud platform; the power supply module provides power to the distributed monitoring nodes; and the local display module displays monitoring data, equipment status, and early warning information in real time.

[0012] The cloud platform stores the received data, pushes early warning notifications to user terminals when early warning conditions are met, and issues downlink control commands to distributed monitoring nodes.

[0013] As a further improvement of the present invention, the frost heave force sensing module includes at least one of a distributed crack surface frost heave force sensor and a single point frost heave force sensor;

[0014] The distributed crack surface frost heave force sensor is suitable for large rock mass cracks and can synchronously monitor the frost heave force distribution at multiple points along the crack length.

[0015] The single-point frost heave force sensor is suitable for indoor rock sample testing and single-point monitoring of narrow seams in the field.

[0016] Both sensors are encapsulated in a composite of low-temperature resistant and moisture-proof epoxy resin and stainless steel, with a protection rating of no less than IP68, making them suitable for cold environments with a temperature range of -40℃ to 30℃.

[0017] As a further improvement of the present invention, the frost heave force sensing module is also equipped with an elastic pressure and self-locking installation mechanism, the installation mechanism comprising:

[0018] The metal base is fixed to the rock surface on both sides of the crack by low-temperature resistant stainless steel screws and has multiple sets of adjustable mounting holes.

[0019] The elastically pressurized support frame is vertically fixed to the inner side of the metal base and has a built-in low-temperature resistant disc spring assembly, which is used to continuously apply a constant pre-tightening contact force to the sensor during the volume deformation process of the freeze-thaw cycle.

[0020] A rotatable fixed sensor bracket is hinged to the end of the support frame, has a rotational adjustment degree of 0° to 90°, and the end is provided with a slot that matches the sensor housing;

[0021] The self-locking installation device is located at the hinge of the rotatable bracket and the support frame. It adopts a ratchet and pawl self-locking structure, which locks the rotational freedom and extension stroke with one key after the sensor is adjusted to the preset angle.

[0022] As a further improvement of the present invention, the power supply module includes a low-temperature lithium iron phosphate battery and a monocrystalline silicon solar charging panel, providing long-term continuous power supply in outdoor environments.

[0023] The intelligent frost heave force analysis algorithm built into the main control module includes the following processing units executed sequentially:

[0024] The data acquisition and preprocessing unit synchronously acquires frost heave force sensing signals and temperature sensor signals at a fixed sampling rate. It performs full-temperature-range segmented temperature compensation, filtering and noise reduction, and dynamic outlier removal from the freeze-thaw process sensing signals on the frost heave force sensing signals, and outputs clean data.

[0025] The freeze-thaw event and feature recognition unit identifies three states—freezing, thawing, and stable period—based on system thresholds, and extracts the maximum frost heave force, duration, number of freeze-thaw cycles, and frost heave force change rate for each freeze-thaw cycle.

[0026] The prediction and risk assessment unit uses an online adaptive ARIMA model to predict the trend of frost heave force changes in the next 25-33 minutes, and uses an SVM classification model trained with data on frost heave failure of rock masses in cold regions to assess the current state with three levels of risk: safe, concerned, and dangerous.

[0027] The multi-level early warning decision and output unit integrates four types of triggering logic: threshold early warning, trend early warning, rate change early warning, and model risk early warning. It generates hierarchical early warning information packages and uploads them to the cloud platform synchronously through a wireless communication module.

[0028] As a further improvement of the present invention, the full-temperature-range segmented temperature compensation divides the temperature range from -40℃ to 30℃ into three intervals, and each interval is corrected using an independent pre-calibrated compensation coefficient. The correction formula is as follows:

[0029] Compensated signal = Original signal ÷ [1 - αi × (20.0℃ - current temperature)];

[0030] Where 20.0℃ is the sensor calibration reference temperature, and αi is the pre-calibration compensation coefficient for the corresponding temperature range; the compensation coefficient α1=0.0015 for -40℃ to -20℃, the compensation coefficient α2=0.0022 for -20℃ to 0℃, and the compensation coefficient α3=0.0018 for 0℃ to 30℃; when the current temperature is below 20.0℃, the denominator is less than 1, which increases the compensated signal relative to the original signal to offset the low reading caused by the decrease in low-temperature sensitivity.

[0031] As a further improvement of the present invention, the dynamic outlier removal adaptively adjusts the judgment threshold according to the current freeze-thaw state: in the frozen or thawing state, 60% of the historical maximum change range is used as the upper limit of the allowable change amount to retain the effective mutation signal of the phase change process; in the stable state, three times the historical change standard deviation is used as the judgment threshold to remove random interference according to the 3σ criterion.

[0032] The largest historical variation was the maximum frost heave force F recorded in the freeze-thaw cycle. max The maximum value of the difference between the initial frost heave force F0 and the time corresponding to the start of freezing;

[0033] The historical variation standard deviation is the standard deviation of the frost heave force data during the most recent hour of the freeze-thaw cycle. When this data is not available, the historical variation standard deviation is 0.5 kPa.

[0034] As a further improvement of the present invention, the online adaptive ARIMA model automatically searches for the optimal order combination and updates the model parameters after each complete freeze-thaw cycle using the AIC criterion within a preset parameter space: p∈[0,2], d∈[0,1], q∈[0,1], with a prediction step count of 3000-4000 steps.

[0035] As a further improvement of the present invention, the input feature vector of the SVM classifier includes the current frost heave force value, the average rate of change of frost heave force V5(t) within 5 minutes, the maximum frost heave force in this cycle, the cumulative number of freeze-thaw cycles, the average of the maximum frost heave forces in the last 10 cycles, and the real-time temperature; before input, Z-score normalization is performed using the mean and standard deviation of the training set; the classifier uses a radial basis function kernel function, with a penalty parameter C of 10, a kernel parameter γ of 0.1, and a convergence tolerance of 10. -3 ;

[0036] The average rate of change of frost heave force V5(t) within 5 minutes is the frost heave force value F at the current moment. c The difference between (t) and the frost heave value 5 minutes ago is divided by the 5-minute time interval.

[0037] The average of the maximum frost heave forces in the last 10 cycles is the arithmetic mean of the maximum frost heave forces in the last 10 complete freeze-thaw cycles.

[0038] The cumulative freeze-thaw cycle count refers to the number of freeze-thaw cycles of soil and rock during the system's operation.

[0039] As a further improvement of this invention, the classification criteria in the SVM classifier are: absolute value of frost heave force, short-term rate of change, historical evolution trend, and auxiliary model results.

[0040] Safety level: When the current real-time frost heave force Fc(t) < 0.3Fs and |dF / dt| ≤ 2Vhist, and the ARIMA prediction trend does not continue to rise, it is judged as a safe level, indicating that the rock mass is currently in a stable state;

[0041] Attention level: When 0.3Fs≤Fc(t)<0.7Fs, or |dF / dt|>2Vhist and the ARIMA prediction trend is upward, it is judged as attention level, indicating that there is potential risk in the rock mass and the monitoring frequency needs to be increased;

[0042] Hazard level: When Fc(t)≥0.7Fs, or the ARIMA prediction trend continues to rise and |dF / dt|>5Vhist, or the SVM classification results output danger for two consecutive sliding windows, it is judged as a hazard level, indicating that the rock mass is about to or has already been damaged, and engineering measures must be taken immediately.

[0043] The material design safety threshold Fs is determined based on the monitored rock mass type, the effective stress area Ae of the sensor, the tensile strength σt of the rock mass, and the fracture toughness KIC. When tensile failure is controlled, Fs = η·σt·Ae; when crack propagation is controlled, Fs = η·KIC·Ae / (πa). 0.5 , where a is the crack half-length, η is a safety reduction factor of 0.3-0.6, and the smaller value among the above calculated values ​​is taken as the material design safety threshold Fs corresponding to the rock mass;

[0044] Vhist is updated to the arithmetic mean of the average rate of change of the most recent M complete freeze processes, where M is 5-15, preferably 10.

[0045] Technical Theme Two

[0046] A method for intelligent monitoring of frost heave force in cold-region geotechnical engineering using the system described in Technical Topic 1, the key of which includes the following steps:

[0047] Step 1: Simultaneously acquire frost heave force and temperature signals, and sequentially perform full-temperature-range segmented temperature compensation, 5-point median filtering, first-order low-pass filtering with a cutoff frequency of 0.1Hz, and dynamic outlier removal based on freeze-thaw state to obtain a clean data sequence.

[0048] Step two: Based on the temperature threshold and the rate of change of frost heave force threshold, determine the freeze-thaw cycle status, identify and record the key characteristic parameters of each complete freeze-thaw cycle, including the current real-time frost heave force F. c (t), average rate of change of frost heave force within 5 minutes V5(t), maximum frost heave force in this cycle F max Cumulative freeze-thaw cycles N cycle The average value of the maximum frost heave force F from the most recent 10 complete freeze-thaw cycles. 10 And real-time ambient temperature T, used for subsequent ARIMA model order updates, dynamic adjustment of outlier thresholds, and SVM risk level assessment;

[0049] Step 3: Use the online optimized ARIMA model to predict the trend of frost heave force in the next 25-33 minutes and determine whether the trend type is continuous increase, stabilization or decline; at the same time, based on the normalized multi-dimensional feature vector, use the SVM classifier to assess the current frost heave risk level.

[0050] Step four involves integrating four types of triggering conditions: threshold warning, trend warning, rate change warning, and model risk warning. The highest level among all triggering conditions is used as the final warning level. A structured warning information package containing time, location, real-time data, warning level, triggering reason, and handling suggestions is generated and simultaneously pushed to the cloud platform and user terminals via wireless network.

[0051] The beneficial effects of this invention are as follows:

[0052] 1. This invention provides an intelligent monitoring system for frost heave force in cold-region geotechnical engineering based on the Internet of Things, which realizes remote, real-time and automated monitoring of frost heave force in rock mass cracks, and can predict the development trend of cracks and issue timely warnings when data is abnormal, which greatly improves the initiative and foresight of engineering safety management.

[0053] 2. This invention provides an intelligent monitoring method for frost heave force in cold-region geotechnical engineering based on the Internet of Things. The method provides accurate results and can realize real-time processing, trend prediction, and intelligent early warning of frost heave force data. Attached Figure Description

[0054] Figure 1 This is a flowchart of the intelligent monitoring method for frost heave force of the present invention;

[0055] Figure 2 This is a partial schematic diagram of the intelligent monitoring system for frost heave force of the present invention;

[0056] Figure 3 This is a partial structural diagram of the detection node of the intelligent monitoring system for frost heave force of the present invention;

[0057] Figure 4 This is an exploded view of the mounting bracket and housing of the frost heave force sensor and wireless communication module in the intelligent frost heave force monitoring system of the present invention.

[0058] Figure 5 This is a partial structural diagram of the mounting bracket for the intelligent monitoring system for frost heave force of the present invention.

[0059] In the diagram: 1: Cloud platform; 6: Wireless network / base station; 701: Power indicator; 702: Working status indicator; 703: Wireless communication status indicator; 704: Signal output negative terminal; 705: Signal output positive terminal; 706, 707: Power supply terminals; 8: Distributed crack surface frost heave force sensor; 9: Single-point frost heave force sensor; 10: Sensor base housing; 11: Sensor housing display screen; 12: Elastic pressure and self-locking mounting mechanism; 1201: Metal base of mounting bracket; 1202: Base screw; 1203: Elastic pressure support frame; 1204: Rotatable fixed sensor bracket; 1205: Self-locking mounting device. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the present invention clearer, the invention will be described clearly and completely below in conjunction with specific embodiments.

[0061] Example 1

[0062] Combination Figure 2 , Figure 4 and Figure 5 The typical deployment method of the IoT-based intelligent monitoring system for frost heave force in cold-region geotechnical engineering for field rock slope monitoring is as follows:

[0063] At the target rock mass crack, the metal base 1201 of the elastic pressure and self-locking installation mechanism 12 is fixed to the stable rock mass surface on both sides of the crack using base screws 1202. The distributed crack surface frost heave force sensor 8 or the single-point frost heave force sensor 9 is then inserted into the dedicated slot of the rotatable fixed sensor bracket 1204, adjusted to an angle perpendicular to the crack surface, and locked with a single key using the self-locking installation device 1205. The low-temperature resistant disc spring assembly within the elastic pressure support frame 1203 ensures that the sensor continuously receives a constant pre-tightening contact force during freeze-thaw cycles.

[0064] The monitoring node includes a sensing module, a signal conversion module, a main control module, a wireless communication module, a power supply module, and a local display module 11.

[0065] The sensing module includes a frost heave force sensing module and a temperature sensing module. The frost heave force sensing module is fixed to both sides of the crack in the rock mass to be monitored or the indoor rock sample, and is used to convert the width change or stress change at the crack into an electrical signal. The temperature sensing module collects the ambient temperature information and converts the ambient temperature information into an electrical signal. The signal conversion module amplifies, filters and performs analog-to-digital conversion on the electrical signal, and outputs a digital signal to the main control module.

[0066] The main control module has a built-in frost heave analysis algorithm to preprocess the collected data, identify the freeze-thaw state, predict trends, and assess risks, and generate early warning decisions; the wireless communication module uploads the processed data to the cloud platform and receives control commands from the cloud platform; the power supply module provides power to the distributed monitoring nodes; and the local display module displays monitoring data, equipment status, and early warning information in real time.

[0067] The cloud platform stores the received data, pushes warning notifications to user terminals when warning conditions are met, and issues downlink control commands to distributed monitoring nodes. Distributed crack surface frost heave force sensors 8 or single-point frost heave force sensors 9 collect crack frost heave force signals in real time. After amplification, filtering, and analog-to-digital conversion by the signal conversion module, the signals are transmitted to the main control module located inside the sensor's base housing 10. The main control module sequentially performs temperature compensation, filtering and noise reduction, freeze-thaw state identification, dynamic outlier removal, ARIMA trend prediction, and SVM risk assessment, generating data packets carrying risk level labels. Under normal conditions, the data packets are periodically uploaded to the cloud platform 1 via the wireless network / base station 6 through the wireless communication module; when a warning is triggered, a warning information packet is automatically added and uploaded immediately, and the cloud platform simultaneously pushes a warning notification to the mobile APP. The sensor's housing display screen 11 displays the frost heave force value, current freeze-thaw state, and equipment operating status in real time for direct reading by on-site inspection personnel. The power management module consists of a solar charging panel and a low-temperature lithium iron phosphate battery, supporting continuous power supply in an environment of -40℃.

[0068] Specifically, this invention designs two sensors for different rock mass crack shapes. For indoor rock mechanics tests of small cracks or single elements, a single-point frost heave force sensor 9 can be used.

[0069] For large cracks, a distributed crack surface frost heave force sensor 8 can be used to detect the frost heave force distribution characteristics on the entire surface, and can be adjusted as needed during use.

[0070] Specifically, the collected data is transmitted via a wireless communication module, and through the communication link formed by the wireless network / base station 6, the uplink monitoring data is transmitted to the remote cloud platform 1. The cloud platform stores, analyzes, and intelligently judges the data, and when instructions need to be issued, it sends downlink control instructions to the field equipment through the opposite path.

[0071] Reference Figure 3 It shows in detail the circuit board port and status indicator structure of the field monitoring node 7.

[0072] This module integrates sensor signal conditioning, main control processing, and wireless communication functions. Terminals 706 and 707 form the power supply terminals and connect to the power supply module; the positive signal output terminal 705 and the negative signal output terminal 704 connect to the frost heave force sensing module. When the module is working, the power indicator 701 illuminates to indicate normal power supply, the operating status indicator 702 displays the self-locking or sampling operating status, and the wireless communication status indicator 703 displays the network connection and data transmission status.

[0073] Reference Figure 5 This invention details the partial structure of the elastic pressure and self-locking installation mechanism 12, a functional unit that ensures continuous tight contact between the sensor and the rock mass / sample during freeze-thaw cycles, preventing displacement and detachment. The structure uses a metal base 1201 as the supporting body, which is fixed to the stable rock mass or test rock sample surface on both sides of the crack by four specially customized base screws 1202. The screws are made of low-temperature resistant stainless steel, suitable for extreme cold environments from -40℃ to 30℃, avoiding low-temperature brittle fracture failure. The base also has multiple sets of adjustable mounting holes to flexibly adapt to installation requirements for cracks of different widths. The elastic pressure support frame 1203 is vertically fixed to the inner side of the metal base 1201, providing a reference support for sensor installation. The support frame has a built-in low-temperature resistant disc spring assembly, which continuously outputs a constant pre-tightening contact force to the sensor during the volume expansion and contraction caused by the water-ice phase change in freeze-thaw cycles, counteracting the volume deformation of the rock mass / sample and preventing contact gaps between the sensor and the measured surface. The rotatable fixed sensor bracket 1204 is hinged to the end of the elastically pressurized support frame 1203, providing a rotational adjustment degree of 0~90° to adapt to cracks and measured surfaces with different inclination angles. The end of the bracket is equipped with a dedicated slot that matches the sensor housing, allowing for quick mounting of a single-point frost heave force sensor 9 or a distributed crack surface frost heave force sensor 8 without the need for additional drilling, significantly improving on-site installation and debugging efficiency. The self-locking installation device 1205 is located at the hinge between the rotatable fixed sensor bracket 1204 and the elastically pressurized support frame 1203, employing a ratchet and pawl self-locking structure. After the sensor is adjusted to the preset angle and the target preload is applied, the rotational freedom and extension stroke of the bracket can be locked with one click, ensuring no loosening or displacement under repeated deformation during freeze-thaw cycles and field vibration conditions, guaranteeing the long-term stability of the sensor's installation posture and contact force.

[0074] The implementation process of the IoT-based intelligent monitoring method for frost heave force in cold-region geotechnical engineering of the present invention is as follows: Figure 1 As shown, this can be achieved through the following steps:

[0075] Step 1:

[0076] The frost heave force sensor was fixed at the position to be measured, and the signal output by the frost heave force sensor and the temperature signal of the high-precision PT100 temperature sensor (resolution 0.1℃) were simultaneously acquired at a fixed sampling rate of 2Hz.

[0077] Based on the temperature information from the temperature sensor, the signal output by the frost heave force sensor is compensated as follows:

[0078] The temperature range of -40℃ to 30℃ is divided into three intervals, and a pre-calibrated piecewise compensation coefficient is used for correction.

[0079] Corrected formula: Compensated signal = Original signal ÷ [1 - αi × (20.0℃ - Current temperature)].

[0080] The coefficients αi are selected according to temperature ranges: α1 = 0.0015 for -40℃ to -20℃, α2 = 0.0022 for -20℃ to 0℃, and α3 = 0.0018 for 0℃ to 30℃. These coefficients are the fitted averages obtained after multiple batches of sensors were calibrated by temperature increases and decreases in a constant temperature chamber.

[0081] After temperature compensation, noise and outliers in the signal are removed, specifically including:

[0082] Adaptive composite filtering for noise reduction: First, a 5-point sliding window mid-range filter is used to remove pulse spike interference caused by ice crystal breakage and instantaneous impact. Then, a first-order low-pass filter with a cutoff frequency of 0.1Hz is used to filter out high-frequency noise generated by electromagnetic interference and mechanical vibration.

[0083] Step 2:

[0084] Dynamic outlier removal based on freeze-thaw process sensing: The outlier threshold is dynamically adjusted based on the current freeze-thaw state.

[0085] The complete freeze-thaw cycle involved in this step includes the entire process from freezing, through thawing, and finally into the stable period. The specific judgment process includes:

[0086] Calculate the frost heave force F at the current moment. c The difference between (t) and the frost heave force value 5 minutes ago is divided by the 5-minute time interval to obtain the average rate of change of frost heave force V5(t) within 5 minutes.

[0087]

[0088] In the formula, the system sampling frequency is 2 Hz, F c (t - 300 s) represents the frost heave force value 5 minutes before the current time.

[0089] Calculate the difference between the current frost heave force value and the frost heave force value 10 minutes ago, and divide the difference by the 10-minute time interval to obtain the average rate of change V of the frost heave force over the 10 minutes. 10 (t); based on V5(t), V 10 The status is determined based on (t) and the current real-time temperature. 1. Status determination during freezing.

[0090] The system is considered to be in a frozen state when all of the following conditions are met:

[0091] The real-time ambient temperature is below -0.5℃, the average rate of change of frost heave force within 5 minutes is greater than 0.05kPa / 5min, and the current system status is not frozen.

[0092] After the determination is completed, the main control module automatically records the start time of this freezing process and the initial frost heave force F0 corresponding to the start time.

[0093] 2. Determination of Stable Period Status

[0094] When the absolute value of the average rate of change of frost heave force within 10 minutes is less than 0.02 kPa / 10 min, the system is determined to have switched to a stable state. The stable state can be directly transitioned from either the freezing or thawing state.

[0095] 3. Determining the state during melting

[0096] The system is considered to be in a melting state when the following conditions are met simultaneously:

[0097] The real-time ambient temperature is above 0.5℃, the average rate of change of frost heave force within 5 minutes is less than -0.05kPa / 5min, and the current system status is frozen.

[0098] After the determination is completed, the main control module performs the following operations in sequence:

[0099] Extract the maximum value F from the frost heave force data of this freezing process, that is, from the start time of the freezing process to the current time. max The freeze-thaw cycle counter N cycle Increment the count value by 1, and set the maximum frost heave force F for this cycle to 1. max The freezing duration and temperature range are stored in the feature library;

[0100] Rolling update of system historical statistical parameters:

[0101] The historical maximum variation range has been updated to the original historical maximum variation range plus the variation range of frost heave force in this cycle (|F). maxThe larger value in -F0|) is used; the historical standard deviation is updated to the standard deviation of the frost heave force data during the most recent 1 hour of stable period;

[0102] The historical average rate of change Vhist is updated to the arithmetic mean of the average rates of change of the most recent M complete freeze processes, where M is preferably 10; when the number of completed cycles is less than 10, the average of all completed cycles is used.

[0103] The average rate of change in a single freezing process is |F max -F0| is the ratio of the freezing duration.

[0104] Status update

[0105] After completing any of the above state determinations, the main control module updates the current system state to the newly determined state, which is used for subsequent dynamic outlier removal, risk assessment, and early warning triggering logic.

[0106] The system continuously monitors: current real-time frost heave force Fc(t), Vinst, short-term instantaneous change rate Vinst (the ratio of the difference between the current real-time frost heave force and the real-time frost heave force 10 seconds ago to 10 seconds ago), maximum frost heave force of this cycle Fmax, cumulative freeze-thaw cycles Ncycle, average maximum frost heave force of the last 10 cycles F10, historical trend slope G10, real-time ambient temperature T, historical standard deviation, historical average change rate, average change rate of frost heave force V5(t) within 5 minutes, and average change rate of frost heave force V within 10 minutes. 10 (t). Where Fc(t) is the frost heave force value of the current sampling point after temperature compensation, filtering and denoising, and removal of dynamic outliers; F10 is the arithmetic mean of the maximum frost heave force of the most recent m complete freeze-thaw cycles. When the cumulative number of freeze-thaw cycles is less than 10, the average of the maximum frost heave forces of all existing complete freeze-thaw cycles is taken. G10 is the least-squares fitting slope of the maximum frost heave force Fmax,j of the most recent m cycles relative to the cycle number j. Using Fref=max(F10, 0.01Fs) as the normalization benchmark, when G10 / Fref is greater than 0.02 / cycle, the historical trend is considered to be upward; when its absolute value is not greater than 0.02 / cycle, the historical trend is considered to be stable; when it is less than -0.02 / cycle, the historical trend is considered to be downward. F10 is used to characterize the recent frost heave force benchmark level, and G10 is used to quantify the historical evolution direction. The removal of dynamic outliers includes the following:

[0107] Frozen / thawing: Use 60% of the historical maximum change as the threshold.

[0108] Stable period: Three times the standard deviation of historical changes is used as the threshold, and random disturbances are eliminated according to the 3σ criterion.

[0109] If three consecutive sampling points exceed the threshold, the sensor self-test is triggered immediately.

[0110] The main control module retains raw data for at least 10 seconds before and after triggering, and maintains the sensor's mechanical installation and stress state unchanged. The signal conversion module sequentially switches the acquisition channel to two built-in reference signals at 20% and 80% of the full-scale range FS via an analog switching circuit. Each level continuously acquires 20 sampling points, calculating the measurement mean and standard deviation. When the relative error of both reference signals is no greater than 1%, and the standard deviation is no greater than 0.005FS, the signal conversion link is considered normal, the sensor input is switched back, and sampling continues; otherwise, the signal conversion link is considered abnormal, fault code E01 is generated, and uploaded to the cloud platform. After switching back to the sensor input, the average value 5 seconds after recovery sampling is compared with the average value 5 seconds before self-test. If the absolute value of the difference is greater than 0.05FS or the signal shows full-scale saturation, fault code E02 is generated. When there is no historical data upon power-on, the following default parameters are used:

[0111] The historical maximum variation range ΔFhist = 10 kPa is the default value for the first start-up. After the first freeze-thaw cycle is completed, it is automatically updated to the maximum value of |Fmax-F0| in the most recent M complete freeze-thaw cycles, with M preferably set to 10.

[0112] The historical variation standard deviation Shist = 0.5 kPa is the default value for the first startup. After completing the 1-hour stabilization period sampling, it is updated to the standard deviation of the most recent 1-hour stabilization period real-time compensation frost heave force sequence.

[0113] The historical average rate of change, Vhist, is 0.02 kPa / 5 min. This parameter is the default value for the first start-up. After the first freeze-thaw cycle is completed, it is updated to the arithmetic mean of the average rates of change of the most recent M complete freezing processes, with M preferably being 10. When the number of completed cycles is less than 10, the average value of all completed cycles is used.

[0114] Freeze-thaw cycle counter N cycle = 0.

[0115] The current state is a stable period.

[0116] Step 3:

[0117] This step uses an autoregressive integral moving average (ARIMA) model to analyze the real-time frost heave force F. c(t) Short-term trend prediction is performed. The initial order of the model is set to (p, d, q) = (1, 1, 1). As monitoring data accumulates, the system combines p, d, and q within the preset parameter range, calculates the Akaike Information Criterion (AIC) value corresponding to each order combination, and selects the order combination with the smallest AIC value as the optimal order of the current model to adapt to the evolution characteristics of frost heave force under different monitoring locations and different lithological conditions.

[0118] Real-time risk level assessment employs a support vector machine (SVM) classification model. The model input features include the current real-time frost heave force F. c (t), average rate of change of frost heave force within 5 minutes V5(t), maximum frost heave force in this cycle F max Cumulative freeze-thaw cycles N cycle The average value of the maximum frost heave force F from the most recent 10 complete freeze-thaw cycles. 10 And the real-time ambient temperature T. Where:

[0119] F 10 This feature is used to characterize the baseline level and historical evolution of frost heave force during recent freeze-thaw cycles at the monitoring point, and serves as an input feature for the SVM classifier in risk level determination, but is not used alone as a warning threshold. Each input feature is Z-score normalized using the training set mean and standard deviation before being input into the classifier.

[0120] The autoregressive integral moving average model is used to predict the short-term trend of frost heave force. Through an online self-optimization mechanism of model order, it dynamically adapts to the evolution of frost heave force in different engineering sites and with different lithologies, avoiding the insufficient adaptability of fixed-parameter models. Its specific workflow is as follows:

[0121] S311 Model Startup Conditions

[0122] When the cumulative data collected by the field monitoring nodes covers at least one complete freeze-thaw cycle, the ARIMA prediction process is automatically initiated. If the data volume is less than one complete cycle, trend prediction and SVM risk assessment are temporarily suspended, and only basic data monitoring and status identification are performed.

[0123] S312 model order online self-optimization

[0124] Based on the frost heave force data from the previous complete freeze-thaw cycle, the system uses Autoregression Intervention (AIC) to automatically search for the optimal model order within a preset parameter space. The autoregressive order *p* ranges from {0, 1, 2}, the difference order *d* ranges from {0, 1}, and the moving average order *q* ranges from {0, 1}. After each complete freeze-thaw cycle, the system automatically updates the model order and fitting coefficients to ensure the model consistently matches the frost heave force evolution characteristics of the current rock mass.

[0125] S313 Short-term frost heave prediction

[0126] The optimized ARIMA model is used to predict the current preprocessed frost heave force sequence, with the prediction step count preferably fixed at 3600 steps. Since the system sampling rate is 2Hz, 3600 steps correspond to a time span of 30 minutes; however, it can also be set in the range of 3000-4000 steps, corresponding to a short-term prediction window of 25-33 minutes, depending on the communication frequency and engineering response requirements.

[0127] To verify the accuracy of the prediction model, the system uses the most recently completed freeze-thaw cycle as the validation window, arranging the first 80% of the sequences as the model fitting segment and the last 20% as the validation segment in chronological order. An ARIMA model is then built using the fitting segment to predict the validation segment. The mean absolute percentage error is calculated as MAPE = (100% / n)Σ|Fi- The expression | / max(|Fi|, ε) is calculated, where Fi and ε are... These represent the measured and predicted values ​​for the validation segment, respectively, where n is the number of sampling points in the validation segment, and ε is set to 0.01Fs to avoid denominator instability when the measured value approaches zero. When the MAPE is not greater than 10% and the overall slope sign of the predicted sequence in the validation segment is consistent with that of the measured sequence, the model is refitted using the entire complete freeze-thaw cycle, and the future trend is output; otherwise, only the real-time monitoring value is output, and the model is re-optimized and validated in the next complete freeze-thaw cycle.

[0128] S314 Trend Classification Judgment

[0129] Based on the predicted frost heave force sequence, and using the frost heave force value at the start of the prediction (i.e., the current real-time frost heave force value) as a benchmark, the trend of frost heave force change over the next 30 minutes is determined according to the following criteria:

[0130] If the ratio of the difference between the maximum value of the predicted sequence and the starting value of the prediction to the starting value of the prediction is greater than or equal to 0.05, and there is no downward inflection point in the predicted sequence, then the predicted trend is determined to be a continuous upward trend.

[0131] If the ratio of the difference between the maximum and minimum values ​​of the predicted sequence to the initial predicted value is less than or equal to 0.02, the predicted trend is determined to be stable.

[0132] If the ratio of the difference between the predicted starting value and the minimum value of the predicted sequence to the predicted starting value is greater than or equal to 0.05, and there is no upward inflection point in the predicted sequence, then the predicted trend is determined to be downward.

[0133] S32 Multidimensional Risk Level Assessment Submodule:

[0134] This submodule uses a support vector machine (SVM) classification model to quantitatively assess the risk of rock mass frost heave damage by integrating multi-dimensional risk features, and outputs a three-level risk rating. The specific implementation method is as follows:

[0135] S321 Model Basics

[0136] The SVM model is trained offline on a host computer or cloud platform. The trained support vectors, dual coefficients, intercepts, and the mean and standard deviation of each feature are then stored in the model parameter table of the field monitoring nodes or the cloud platform. The field monitoring nodes only perform model inference and do not perform online training. Training samples are generated in 5-minute data windows, each containing 600 sampling points, with an adjacent window step size of 120 sampling points (1 minute). At the end of each window, x=[Fc / Fs, Vinst / Vhist, Fmax / F10, G10, Ncycle, T] is extracted. Window labels are determined according to the explicit threshold rules in S322 of this specification, and the highest risk level appearing within the window is taken as the window label. When there are field-confirmed records of crack propagation, component cracking, or rock mass failure, the corresponding window is directly marked as dangerous. Thus, the feature calculation, window division, and label generation of training samples all have defined rules.

[0137] The input features are Z-score normalized: x'=(x-μ) / s, where μ and s are the mean and standard deviation of the corresponding feature in the training set, respectively. The SVM classifier uses a one-to-one multi-class strategy and a radial basis function kernel function K(xi, xj)=exp[-γ||xi-xj||²], with a penalty parameter C=10, a kernel parameter γ=0.1, a maximum number of iterations of 10000, and a convergence tolerance of 10. -3 After the model completes training and validation on the host computer, the parameters are directly fixed without setting a field judgment step of "writing to the field node only after reaching a certain accuracy". The field node loads the support vectors, coefficients, intercepts, μ and s after the model file passes the validation and performs classification according to the fixed parameters.

[0138] Table 1 SVM Training and Deployment Parameters 1.

[0140] 2. S322 Risk Level Classification Basis

[0141] The system combines the absolute value of frost heave force, short-term change rate, historical evolution trend, and auxiliary model results to classify the risk of rock mass frost heave failure into the following three levels. The material design safety threshold Fs is determined according to the failure control mode of the monitored rock mass: for rock mass controlled by tensile failure, Fs,t=η·σt·Ae; for rock mass controlled by crack propagation, Fs,K=η·KIC·Ae / (πa)0.5 When both σt and KIC are available, the main control module takes min(Fs,t,Fs,K) as Fs for that measuring point; if only one parameter is available, the corresponding calculated value is used. σt can be obtained according to the tensile strength test of rock blocks in GB / T 50266-2013 "Standard for Test Methods of Engineering Rock Mass", or the value of the reviewed exploration test report can be used; KIC can be obtained according to the CCNBD test method or the semi-circular bending SCB test method recommended by the International Society for Rock Mechanics and Rock Engineering (ISRM); Ae is determined by the sensor's stress structure dimensions and calibration records; a is determined by half of the equivalent crack length obtained by crack image measurement, crack gauge measurement, or on-site measurement; η is selected in the range of 0.3-0.6 according to the importance and safety level of the project, with smaller values ​​for higher project importance. When deploying each measuring point, the above parameters and the calculated Fs are written into the parameter table.

[0142] Table 2 Material Design Safety Threshold Parameters and Acquisition Methods

[0143]

[0144] Step 4

[0145] This step integrates the results of the multi-dimensional analysis above to form the final engineering early warning decision. The early warning does not rely on a single condition, but rather integrates four parallel triggering logics. All thresholds and level classifications are completely consistent with the algorithm command flow:

[0146] Threshold warning: Triggered when the current frost heave force exceeds 0.7 times the material design safety threshold preset according to the rock mass mechanical properties, corresponding to a level one red warning;

[0147] Trend Warning: When the ARIMA model predicts a continuous upward trend and the SVM risk assessment level is "Attention", a Level 2 Yellow Warning is triggered; when the predicted trend is a continuous upward trend and the risk level is "Dangerous", a Level 1 Red Warning is triggered.

[0148] Rate of sudden change warning: Triggered when the absolute value of the instantaneous rate of change of frost heave exceeds 5 times the historical average rate of change, corresponding to a level one red warning;

[0149] Model risk warning: When the SVM risk assessment level is dangerous for two consecutive sliding windows, and the current real-time frost heave force Fc(t) is not less than 0.5Fs or |dF / dt| exceeds 3Vhist, a level one red warning is triggered.

[0150] Once any warning trigger condition is met, the system immediately determines the final warning level (taking the highest warning level corresponding to all trigger conditions) and automatically generates a structured warning information package containing a timestamp, node location, real-time monitoring data, final warning level, trigger reason, and engineering handling suggestions. This information is simultaneously broadcast in real time to local displays, cloud monitoring platforms, and authorized user terminals via a wireless communication module, achieving multi-channel and full-coverage delivery of warning information.

[0151] According to the preset communication protocol, the key data generated in steps 1-3, including the current real-time frost heave force Fc(t), temperature T, freeze-thaw state, risk level, and predicted trend, along with necessary metadata such as time and device ID, are encapsulated into a core data packet. If an early warning is triggered, a complete early warning information packet is attached to the core data packet and uploaded to the cloud platform periodically or on an event-triggered basis via a wireless network for long-term storage and in-depth analysis.

[0152] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent monitoring system for frost heave force in cold-region geotechnical engineering based on the Internet of Things, characterized in that, It includes: distributed monitoring nodes and a remote cloud platform; The monitoring node includes: a sensing module, a signal conversion module, a main control module, a wireless communication module, a power supply module, and a local display module; The sensing module includes a frost heave force sensing module and a temperature sensing module. The frost heave force sensing module is fixed to both sides of the crack in the rock mass to be monitored or the indoor rock sample, and is used to convert the width change or stress change at the crack into an electrical signal. The temperature sensing module collects the ambient temperature information and converts the ambient temperature information into an electrical signal. The signal conversion module amplifies, filters, and performs analog-to-digital conversion on the electrical signal, and outputs a digital signal to the main control module. The main control module has a built-in frost heave analysis algorithm to preprocess the collected data, identify the freeze-thaw state, predict trends, and assess risks, and generate early warning decisions; the wireless communication module uploads the processed data to the cloud platform and receives control commands from the cloud platform; the power supply module provides power to the distributed monitoring nodes; and the local display module displays monitoring data, equipment status, and early warning information in real time. The cloud platform stores the received data, pushes early warning notifications to user terminals when early warning conditions are met, and issues downlink control commands to distributed monitoring nodes.

2. The system according to claim 1, characterized in that, The frost heave force sensing module includes at least one of a distributed crack surface frost heave force sensor and a single point frost heave force sensor.

3. The system according to claim 1, characterized in that, The frost heave force analysis algorithm built into the main control module includes the following processing units executed sequentially: The data acquisition and preprocessing unit synchronously acquires frost heave force sensing signals and temperature sensor signals at a fixed sampling rate. It performs full-temperature-range segmented temperature compensation, filtering and noise reduction, and dynamic outlier removal from the freeze-thaw process sensing signals on the frost heave force sensing signals, and outputs clean data. The freeze-thaw event and feature recognition unit identifies three states—freezing, thawing, and stable period—based on system thresholds, and extracts the maximum frost heave force, duration, number of freeze-thaw cycles, and frost heave force change rate for each freeze-thaw cycle. The prediction and risk assessment unit uses an online adaptive ARIMA model to predict the trend of frost heave force changes in the next 25-33 minutes, and uses an SVM classification model trained with data on frost heave failure of rock masses in cold regions to assess the current state with three levels of risk: safe, concerned, and dangerous. The multi-level early warning decision and output unit integrates four types of triggering logic: threshold early warning, trend early warning, rate change early warning, and model risk early warning. It generates hierarchical early warning information packages and uploads them to the cloud platform synchronously through a wireless communication module.

4. The system according to claim 3, characterized in that, The full-temperature-range segmented temperature compensation divides the temperature range from -40℃ to 30℃ into three intervals, each interval is corrected using an independent pre-calibrated compensation coefficient, and the correction formula is as follows: Compensated signal = Original signal ÷ [1 - αi × (20.0℃ - current temperature)]; Where 20.0℃ is the sensor calibration reference temperature, αi is the pre-calibration compensation coefficient for the corresponding temperature range; the compensation coefficient α1=0.0015 for -40℃ to -20℃, the compensation coefficient α2=0.0022 for -20℃ to 0℃, and the compensation coefficient α3=0.0018 for 0℃ to 30℃.

5. The system according to claim 3, characterized in that, The dynamic outlier removal adaptively adjusts the judgment threshold according to the current freeze-thaw state: when frozen or thawing, 60% of the historical maximum change is used as the upper limit of the allowable change to retain the effective abrupt change signal of the phase transition process; when in a stable state, three times the standard deviation of the historical change is used as the judgment threshold to remove random interference according to the 3σ criterion. The largest historical variation was the maximum frost heave force F recorded in the freeze-thaw cycle. max The maximum value of the difference between the initial frost heave force F0 and the time corresponding to the start of freezing; The historical variation standard deviation is the standard deviation of the frost heave force data during the most recent hour of the freeze-thaw cycle. When this data is not available, the historical variation standard deviation is 0.5 kPa.

6. The system according to claim 3, characterized in that, After each complete freeze-thaw cycle, the online adaptive ARIMA model uses the AIC criterion to automatically search for the optimal order combination within a preset parameter space: p∈[0,2], d∈[0,1], q∈[0,1], and updates the model parameters, with a prediction step count of 3000-4000 steps.

7. The system according to claim 3, characterized in that, The input feature vector of the SVM classifier includes the current frost heave force value, the average rate of change of frost heave force V5(t) within 5 minutes, the maximum frost heave force in the current cycle, the cumulative number of freeze-thaw cycles, the average of the maximum frost heave force in the last 10 cycles, and the real-time temperature. Before input, Z-score normalization is performed using the training set mean and standard deviation. The classifier uses a radial basis function kernel, with a penalty parameter C of 10, a kernel parameter γ of 0.1, and a convergence tolerance of 10. -3 ; The average rate of change of frost heave force V5(t) within 5 minutes is the frost heave force value F at the current moment. c The difference between (t) and the frost heave value 5 minutes ago is divided by the 5-minute time interval. The average of the maximum frost heave forces in the last 10 cycles is the arithmetic mean of the maximum frost heave forces in the last 10 complete freeze-thaw cycles. The cumulative freeze-thaw cycle count refers to the number of freeze-thaw cycles of soil and rock during the system's operation.

8. The system according to claim 3, characterized in that, The criteria for classifying levels in the SVM classifier are: absolute value of frost heave force, short-term rate of change, historical evolution trend, and auxiliary model results. Safety level: When the current real-time frost heave force Fc(t) < 0.3Fs and |dF / dt| ≤ 2Vhist, and the ARIMA prediction trend does not continue to rise, it is judged as a safe level, indicating that the rock mass is currently in a stable state; Attention level: When 0.3Fs≤Fc(t)<0.7Fs, or |dF / dt|>2Vhist and the ARIMA prediction trend is upward, it is judged as attention level, indicating that there is potential risk in the rock mass and the monitoring frequency needs to be increased; Hazard level: When Fc(t)≥0.7Fs, or the ARIMA prediction trend continues to rise and |dF / dt|>5Vhist, or the SVM classification results output danger for two consecutive sliding windows, it is judged as a hazard level, indicating that the rock mass is about to or has already been damaged, and engineering measures must be taken immediately. The material design safety threshold Fs is determined based on the monitored rock mass type, the effective stress area Ae of the sensor, the tensile strength σt of the rock mass, and the fracture toughness KIC. When tensile failure is controlled, Fs = η·σt·Ae; when crack propagation is controlled, Fs = η·KIC·Ae / (πa). 0.5 , where a is the crack half-length, η is a safety reduction factor of 0.3-0.6, and the smaller value among the above calculated values ​​is taken as the material design safety threshold Fs corresponding to the rock mass; Vhist is updated to the arithmetic mean of the average rate of change of the most recent M complete freeze processes, where M is 5-15.

9. A monitoring method based on the IoT-based intelligent monitoring system for frost heave force in cold-region geotechnical engineering as described in any one of claims 1-8, characterized in that, Includes the following steps: Step 1: Simultaneously acquire frost heave force and temperature signals, and sequentially perform full-temperature-range segmented temperature compensation, 5-point median filtering, first-order low-pass filtering with a cutoff frequency of 0.1Hz, and dynamic outlier removal based on freeze-thaw state to obtain a clean data sequence. Step two: Based on the temperature threshold and the rate of change of frost heave force threshold, determine the freeze-thaw cycle status, identify and record the key characteristic parameters of each complete freeze-thaw cycle, including the current real-time frost heave force F. c (t), average rate of change of frost heave force within 5 minutes V5(t), maximum frost heave force in this cycle F max Cumulative freeze-thaw cycles N cycle The average value of the maximum frost heave force F from the most recent 10 complete freeze-thaw cycles. 10 And real-time ambient temperature T, used for subsequent ARIMA model order updates, dynamic adjustment of outlier thresholds, and SVM risk level assessment; Step 3: Use the online optimized ARIMA model to predict the trend of frost heave force in the next 25-33 minutes and determine whether the trend type is continuous increase, stabilization or decline; at the same time, based on the normalized multi-dimensional feature vector, use the SVM classifier to assess the current frost heave risk level. Step four involves integrating four types of triggering conditions: threshold warning, trend warning, rate change warning, and model risk warning. The highest level among all triggering conditions is used as the final warning level. A structured warning information package containing time, location, real-time data, warning level, triggering reason, and handling suggestions is generated and simultaneously pushed to the cloud platform and user terminals via wireless network.

Images