A water mist removal method and system for an unmanned perception device

Abstract

The application discloses a water mist removal method and system of an unmanned sensing device, relates to detection control, and comprises the following steps: collecting temperature data and humidity data of a vehicle-mounted environment in real time through at least two temperature and humidity sensors; performing filtering processing on the collected temperature data and humidity data, and calculating a historical temperature change rate to form a feature vector; inputting the feature vector into a pre-trained lightweight neural network model, performing reasoning, and outputting a water mist level of a current environment, wherein the water mist level comprises four levels of safety, early warning, mildness and severity; determining corresponding air supply temperature parameters and fan rotating speed parameters through a preset mapping relationship table according to the water mist level; calculating a power control signal of a PTC heating sheet by using a fuzzy PID algorithm according to the air supply temperature parameters; and simultaneously generating a rotating speed control signal of the fan according to the fan rotating speed parameters. In view of the fact that a traditional passive demisting mode fails within several minutes in a high-humidity saturated underground environment, the application improves the active demisting effect through hierarchical control.

Description

Technical Field

[0001] This application relates to the field of detection and control, and in particular to a method and system for removing water mist from an unmanned sensing device. Background Technology

[0002] With the rapid development of intelligent and automated technologies, autonomous driving technology has been gradually applied to special operating environments such as underground mines, ports, and tunnels. Autonomous driving systems rely on various sensing devices such as cameras, lidar, and millimeter-wave radar to acquire environmental information in real time, ensuring safe vehicle operation. However, underground mine environments have uniquely harsh conditions: poor ventilation leads to poor air circulation, humidity is often close to saturation (95%–100% RH), temperature fluctuates drastically, and flammable and explosive gases such as methane are present, placing strict explosion-proof requirements on electrical equipment. In this high-humidity, saturated environment, water vapor easily condenses on the surface of sensing devices, severely affecting the sensing accuracy and reliability of the sensors and threatening the operational safety of autonomous vehicles.

[0003] Existing water mist removal technologies for autonomous driving sensing devices are mainly divided into two categories: passive defogging and active defogging.

[0004] Passive defogging technology mainly reduces or delays water mist formation by modifying the surface of materials, including: (1) hydrophobic / hydrophilic coating technology, such as nano-silica hydrophobic coating that makes it difficult for water droplets to adhere, or polyvinyl alcohol hydrophilic coating that makes water mist spread into a transparent water film; (2) anti-fogging film technology, attaching a polymer film containing surfactants to the sensor surface to reduce the surface tension of water; (3) micro-nano structure surface technology, changing the contact angle of water droplets by forming micron / nano-scale rough structures through laser etching.

[0005] Active defogging technology dynamically removes water mist by inputting external energy, including: (1) electric heating defogging, which integrates an ITO transparent conductive film or metal heating wire on the sensor surface and evaporates water mist by generating heat through current; (2) airflow defogging, which uses a fan or compressed air to blow on the sensor surface, often combined with heating to form hot air defogging; (3) thermoelectric cooling technology, which controls the surface temperature through a semiconductor module to prevent condensation; and (4) ultrasonic vibration, which uses high-frequency vibration to remove water mist particles from the surface.

[0006] However, existing technologies have serious limitations in downhole environments:

[0007] First, passive defogging technology is essentially ineffective in the high-humidity saturated environment underground. Due to poor ventilation underground, humidity remains above 95% for extended periods. Traditional passive technologies such as hydrophobic / hydrophilic coatings and antifogging films can only maintain their effectiveness for a few minutes under such extreme conditions. Water mist will quickly recondense on the sensor surface, failing to meet the requirements for continuous operation.

[0008] Secondly, active demisting technology is severely constrained by underground explosion-proof requirements. Underground environments may contain flammable and explosive gases such as methane, requiring electrical equipment to meet intrinsically safe explosion-proof standards. Traditional continuous high-power heating solutions (typically requiring over 200W) are simply unusable in the underground environment. Existing active demisting systems lack intelligent control capabilities, failing to dynamically adjust power based on actual water mist conditions. This results in either insufficient power for effective demisting or continuous high-power operation violating explosion-proof regulations.

[0009] Furthermore, existing technologies lack specific designs for the unique underground environment. Underground temperatures fluctuate dramatically, especially when vehicles move from high-temperature working faces to low-temperature roadways; the sudden temperature drop can easily trigger explosive water mist condensation. Existing systems have slow response times and cannot handle such sudden situations promptly. Simultaneously, existing systems do not consider the impact of high humidity underground environments on control strategies; simple on / off control or linear control cannot adapt to the complex and ever-changing underground environment. Summary of the Invention

[0010] To address the issue that traditional passive defogging methods fail within minutes in high-humidity saturated environments underground, this application provides a water mist removal method and system for unmanned sensing devices. Based on a four-level intelligent control water mist removal method, the active defogging effect is improved through graded control while meeting underground explosion-proof safety requirements.

[0011] One aspect of this application provides a water mist removal method for an unmanned driving sensing device, used in underground mining environments, comprising: S1, real-time acquisition of temperature and humidity data of the vehicle environment using at least two temperature and humidity sensors; wherein the temperature and humidity sensors are installed at the front of the vehicle to avoid the vehicle's own heat dissipation area; S2, filtering the acquired temperature and humidity data and calculating the historical temperature change rate. The process involves: S3, generating a feature vector containing current temperature, current humidity, and historical temperature change rate; S4, inputting the feature vector into a pre-trained lightweight neural network model for inference, outputting the current environmental water mist level, which includes four levels: safe, warning, light, and heavy; S5, determining the corresponding air supply temperature and fan speed parameters based on the water mist level using a preset mapping table, where the air supply temperature ranges from 0 to 40℃ and the fan speed ranges from 0 to 5000 r / min; S6, calculating the power control signal for the PTC heating element using a fuzzy PID algorithm based on the air supply temperature parameter, and simultaneously generating a fan speed control signal based on the fan speed parameter; and S7, controlling the PTC heating element and fan through an intelligent control unit, delivering the generated hot air through pipes to the base of the lidar and camera sensors, and then delivering the hot air to the sensor surface through the air outlet of the sensor base to remove water mist.

[0012] Among them, the fuzzy PID algorithm is an intelligent control method that combines fuzzy control theory with classical PID control. It includes three control elements: proportional (P), integral (I), and derivative (D). A PTC (Positive Temperature Coefficient) heating element is a special ceramic semiconductor heating element. The resistance of PTC material increases sharply with temperature. After reaching the Curie temperature, the resistance increases exponentially, automatically limiting current and power to achieve self-temperature control. As the core heating element, the PTC heating element's power is precisely controlled through the fuzzy PID algorithm, working in conjunction with a fan to generate controllable hot air from 0-40℃, achieving intelligent removal of water mist from the surfaces of lidar and cameras.

[0013] Furthermore, the temperature data collection range is -40℃ to 85℃. In particular, the underground environment experiences extremely drastic temperature changes. When unmanned vehicles rapidly travel from high-temperature mining faces (reaching 40-50℃) to ventilation tunnels or wellheads, the ambient temperature can plummet below 0℃, even dropping below -20℃ in winter. This extreme temperature difference is a major cause of explosive water mist condensation. Setting up a wide data collection range of -40℃ to 85℃ ensures the system can capture temperature data under various extreme underground conditions, providing a complete data foundation for subsequent water mist risk assessment.

[0014] The humidity data collection range is 0 to 100% RH; in particular, poor ventilation underground leads to a large amount of water vapor retention, with humidity remaining at a saturation state above 95% for extended periods. Traditional systems often neglect precise measurement within the high humidity saturation range, but it is precisely the minute changes within this narrow range of 95%–100% RH that determine whether water mist will condense instantly in large quantities. Full-range data acquisition, especially precise monitoring of the high humidity saturation state, is key to identifying the risk of water mist condensation.

[0015] The sampling frequency is once per second.

[0016] Furthermore, S2 filters the collected temperature and humidity data and calculates the historical temperature change rate. This generates a feature vector containing current temperature, current humidity, and historical temperature change rates. The process involves filtering the temperature and humidity data using a Kalman filter algorithm. Furthermore, the strong electromagnetic interference and mechanical vibrations generated by large underground mining equipment lead to significant noise in the raw sensor data. Kalman filtering, based on the system state equation and observation equation, can estimate the true temperature and humidity values ​​in real time, effectively suppressing random noise and improving data reliability.

[0017] The filtered data was then smoothed a second time using a moving average method with a window length of 10 sampling points to eliminate high-frequency noise generated by downhole equipment. The frequent start-up and shutdown of downhole equipment generates high-frequency noise that causes significant data fluctuations. The 10-sampling-point moving average window (corresponding to 10 seconds) effectively filters out these transient interferences while preserving the trend changes in temperature and humidity, preventing the system from frequently switching operating states due to noise misjudgment.

[0018] The filtered humidity data is subjected to saturation detection. When the humidity value at 5 consecutive sampling points exceeds 95%RH, it is marked as humidity saturation. When the humidity exceeds 95%RH for 5 consecutive seconds, it indicates that the environment has entered a high-risk state for water vapor condensation. At this time, even a slight drop in temperature will cause a large amount of water vapor to be generated instantly.

[0019] The historical temperature change rate is obtained by performing a least-squares linear fit on the temperature data within the sliding window. When humidity is saturated, the length of the sliding window is shortened from 10 sampling points to 5 sampling points to improve sensitivity to temperature gradient changes. Specifically, water mist condensation depends not only on the current temperature and humidity but also on the rate of temperature decrease. By calculating the rate of temperature change through least-squares fitting, the system can predict water mist condensation trends. For example, when a temperature decrease at a rate of -2℃ / min and humidity approaches saturation, defogging can be initiated in advance for preventative control.

[0020] The current temperature T, current humidity RH, and historical temperature change rate are used. The feature vector is obtained by combining the features. .

[0021] Further, in S3, the feature vectors are input into a pre-trained lightweight neural network model for inference, outputting the current environmental water mist level. The water mist level includes four levels: safe, warning, light, and heavy. This involves: constructing a lightweight neural network model based on TinyLSTM; simulating an underground environment in a temperature and humidity test chamber, collecting temperature and humidity data under condensation conditions as a training set, with a temperature range of -20℃ to 60℃ and a humidity range of 10%RH to 100%RH; training the model using the cross-entropy loss function and Adam optimizer based on the training set, using the F1 score as the model evaluation metric to obtain the trained TinyLSTM model; using TensorRT to lightweight the trained TinyLSTM model to accelerate the model inference process, obtaining a pre-trained lightweight neural network model; and utilizing the real-time collected and pre-processed feature vectors... Input a pre-trained lightweight neural network model, and through forward propagation, output the probability distribution of four water mist levels. Select the level with the highest probability as the water mist level of the current environment.

[0022] In particular, downhole water mist condensation is a complex nonlinear temporal process that traditional threshold judgments or simple rules cannot accurately describe. There is a complex coupling relationship between temperature, humidity, and the rate of temperature change, especially in the high humidity saturation range of 95%–100% RH, where even small temperature changes can lead to drastic changes in the water mist state. TinyLSTM, as a lightweight long short-term memory network, can: capture temporal dependencies: water mist condensation depends not only on the current state but also on historical temperature and humidity trends. The LSTM's memory mechanism can learn the pattern characteristics before and after sudden changes in downhole temperature. Handle nonlinear mappings: the relationship between downhole environmental parameters and water mist condensation risk is highly nonlinear, and neural networks can automatically learn this complex mapping. Meet vehicle deployment requirements: compared to standard LSTM, TinyLSTM significantly reduces the number of parameters and computational load, making it suitable for real-time operation on computationally limited vehicle-mounted equipment.

[0023] Furthermore, a lightweight neural network model based on TinyLSTM is constructed, including setting the input layer dimension to 3 to receive feature vectors. The LSTM hidden layer is set to have 32 nodes and a time window length of 10 to acquire the rapid condensation process caused by small temperature changes under high humidity saturation conditions in the downhole environment. The output layer is set to have 4 nodes, corresponding to four water mist levels: safety, warning, light, and heavy. The heavy level is used to identify extreme condensation conditions where humidity exceeds 95% and temperature drops rapidly.

[0024] Furthermore, S4, the mapping table, includes: when the water mist level is safe, the supply air temperature is set to 0℃ and the fan speed to 0r / min; however, since there are flammable and explosive gases such as methane underground, any unnecessary electrical operation increases the safety risk. When the environment is far from the dew point, setting 0℃ / 0r / min achieves complete shutdown.

[0025] When the water mist level is at the warning level, the air supply temperature is set to 20℃ and the fan speed to 1200 rpm. The typical downhole ambient temperature is 10-15℃. When approaching the dew point, the sensor surface temperature needs to be 3-5℃ higher than the dew point to prevent condensation. Setting the air supply temperature to 20℃, considering pipe heat loss and air mixing, keeps the sensor surface within a safe range of 2-3℃ above the dew point. This temperature rise effectively prevents condensation without triggering excessively high PTC power. Downhole ventilation is poor, and airflow is slow. 1200 rpm (approximately 2 m / s wind speed) can create a weak but continuous airflow over the sensor surface, breaking the stagnant air layer and preventing localized humidity accumulation. This speed has been tested in downhole operations and effectively disturbs the boundary layer without generating excessive noise that could affect downhole communication.

[0026] When the water mist level is light, set the supply air temperature to 30℃ and the fan speed to 3000 rpm. Specifically, once water mist has formed, it's crucial to evaporate the water droplets quickly. Based on mass and heat transfer calculations, 30℃ hot air can completely evaporate micron-sized water droplets within 10–15 seconds without exceeding the PTC power safety threshold of 100W. This temperature represents the optimal balance between evaporation rate and power consumption in the high-humidity environment of the well.

[0027] When the water mist level is severe, the supply air temperature is set at 40℃ and the fan speed at 5000 rpm. Specifically, the explosion-proof standard for underground wells requires the surface temperature to not exceed 135℃. Considering the thermal resistance and heat transfer coefficient of the PTC heating element, the PTC surface temperature corresponding to a 40℃ supply air temperature is approximately 120℃, leaving a safety margin of 15℃. This is the highest supply air temperature while meeting explosion-proof requirements.

[0028] Furthermore, S4, based on the water mist level, through a preset mapping table, determines the corresponding air supply temperature parameters and fan speed parameters, including: when the humidity value of 5 consecutive sampling points exceeds 95%RH, the air supply temperature corresponding to each water mist level in the mapping table is increased by 2℃ and the fan speed is increased by 10% to compensate for heat loss and water vapor retention caused by poor ventilation in the well.

[0029] When a water mist level is detected to jump directly from the safe level to light or heavy within a preset time, the heavy level demisting parameters are immediately executed, skipping the intermediate transition stage, in response to water mist condensation caused by a sudden drop in underground temperature. In particular, unmanned vehicles often encounter extreme temperature changes when operating underground: when entering a transport roadway (10-15℃) from a high-temperature coal face (40-50℃), the temperature drop rate can reach 5℃ / min; when passing through local ventilation openings, direct cold air can cause a sudden drop in local temperature of more than 10℃. This sudden change causes the water mist level to jump directly from safe to light or even heavy within 10-20 seconds, while the traditional step-by-step temperature increase strategy (safe → warning → light → heavy) takes 30-40 seconds, far too slow to keep up with the water mist condensation rate. In this application, when a level jump is detected (such as jumping from safe to mild within 2 seconds), it indicates that a drastic change has occurred in the environment. At this time, the strongest defogging parameters of 40℃ / 5000r / min are immediately executed; the intermediate transition is skipped to save 20 to 30 seconds of response time; it is better to have a short period of "over-defogging" than to prevent a large amount of water mist from accumulating and affecting safety.

[0030] For light and heavy levels, a minimum continuous defogging time is set to prevent repeated condensation of water vapor in the high-humidity underground environment. Specifically, the underground environment differs from the surface environment, exhibiting unique phenomena of repeated water vapor condensation: poor ventilation prevents the timely removal of evaporated water vapor, creating a high-humidity microenvironment around the sensor; low tunnel wall temperatures cause water vapor to condense and then re-evaporate, resulting in humidity fluctuations; and dust generated during mining operations acts as condensation nuclei, accelerating water vapor reformation. In this application, for light levels: a minimum duration of 60 seconds ensures the microenvironment humidity drops below 85%; for heavy levels: a minimum duration of 120 seconds ensures complete removal of accumulated water vapor. This continuous defogging avoids the problems caused by frequent system start-ups and shutdowns: preventing the water vapor "clearing-condensation-re-clearing" cycle; reducing the number of thermal cycles of the PTC heating element, and extending its service life.

[0031] Furthermore, in step S5, based on the supply air temperature parameter, a fuzzy PID algorithm is used to calculate the power control signal for the PTC heating element; simultaneously, based on the fan speed parameter, a fan speed control signal is generated, including: obtaining the current actual temperature. and air supply temperature Calculate temperature deviation And calculate the rate of change of deviation. .

[0032] Based on the water mist level output in step S3, set the control parameters for the fuzzy PID algorithm: when the water mist level is safe, set... The zero-parameter setting ensures that even minor disturbances will not produce output, avoiding the oscillation problem near zero in traditional PID controllers. When the water mist level is a warning level, the setting is... This parameter combination allows the system to reach the target temperature within 20-30 seconds, with power stabilizing at 30-50W. When the water mist is at a light level, set... To ensure rapid defogging, the system enhances integral action to eliminate steady-state errors and ensures target temperature is reached even in high-humidity environments. This parameter combination can remove light water mist within 15–20 seconds, with a power range of 50–100W. For heavy water mist conditions, the setting is [not specified]. The output power is limited to 100 to 150W; among them, the strong integral action can eliminate deviations even under harsh operating conditions.

[0033] Based on temperature deviation Deviation change rate Based on the set PID control parameters, the initial power of the PTC heating element is calculated; the initial power of the PTC heating element is then subjected to downhole explosion-proof safety constraints to generate the final output power. Based on the fan speed parameters and the PTC heating element power Pout, a synchronized fan speed control signal is generated.

[0034] In particular, the temperature and humidity changes in the downhole environment exhibit significant nonlinearity and time-varying characteristics: under high humidity saturation, the heat transfer coefficient changes nonlinearly with humidity; the resistance-temperature characteristics of the PTC heating element vary greatly in different temperature ranges; downhole airflow disturbances cause the dynamic characteristics of the system to change continuously; traditional fixed-parameter PID cannot adapt to this complex working condition, while fuzzy PID dynamically adjusts the control parameters by water mist level, realizing adaptive control of the nonlinear system.

[0035] Furthermore, the initial power of the PTC heating element is subjected to downhole explosion-proof safety constraints to generate the final output power. This includes: setting the output power when the initial power of the PTC heating element is greater than 150W. To ensure that the surface temperature of the PTC heating element does not exceed the explosion-proof safety threshold; when the output power is maintained at 150W and the temperature deviation is within 3 consecutive sampling cycles. If the condition is abnormal, the output power Pout will be reduced to 100W and an alarm signal will be issued to prevent continuous high power output due to sensor failure.

[0036] Another aspect of this application provides a water mist removal system for an unmanned sensing device used in underground mining environments, comprising: a data acquisition unit including at least two temperature and humidity sensors mounted at the front of the vehicle to avoid the vehicle's own heat dissipation area, for real-time acquisition of temperature and humidity data of the vehicle environment; and an intelligent computing unit that performs Kalman filtering and moving average processing on the acquired temperature and humidity data, and calculates the historical temperature change rate. , forming feature vectors Deploy a lightweight neural network model based on TinyLSTM, input feature vectors into the model for inference, and output water mist level;

[0037] The air supply temperature and fan speed parameters are determined based on the water mist level using a mapping table; a fuzzy PID algorithm is used to calculate the power control signal for the PTC heating element and generate the fan speed control signal; the intelligent control unit, including the PTC heating element, limits the maximum power of the PTC heating element to 150W to meet the explosion-proof requirements in the well; heating is performed according to the power control signal; the fan generates airflow according to the speed control signal; and the purging unit, including an air supply duct, has its inlet end connected to the intelligent control unit, PTC heating element, and fan for delivering hot air.

[0038] Compared to existing technologies, the advantages of this application are:

[0039] To address the issues of traditional passive defogging technologies (coatings, anti-fogging films) failing within minutes and rapid recondensation of water mist in high-humidity saturated environments underground, and the inapplicability of traditional continuous high-power heating solutions due to underground gas explosion-proof requirements, this application addresses these problems. It involves real-time acquisition of temperature and humidity data, calculation of temperature change rate to form a feature vector, inputting this vector into a lightweight neural network model to determine the water mist level (safe, warning, mild, severe). Based on the water mist level mapping, it determines the air supply temperature and fan speed parameters, employs a fuzzy PID algorithm to precisely control the PTC heating element power (limited to 150W), activating heating only when necessary and avoiding power overshoot. The generated hot air is delivered to the sensor surface through pipes and the sensor base's air outlet to remove water mist.

[0040] This application, while meeting the requirements for explosion-proof safety in underground wells, accurately identifies the risk of water mist condensation through an intelligent hierarchical control strategy and dynamically adjusts the demisting parameters. This not only solves the technical problem of rapid water mist condensation in high-humidity underground environments, but also avoids the safety hazards and energy waste caused by continuous high-power heating, ensuring that the lidar and camera sensing devices of unmanned vehicles maintain a clear field of vision in harsh underground environments. Attached Figure Description

[0041] This application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0042] Figure 1 This is an exemplary flowchart illustrating a water mist removal method for an unmanned driving sensing device according to some embodiments of this application;

[0043] Figure 2 This is a schematic diagram of the installation of a temperature and humidity sensor according to some embodiments of this application;

[0044] Figure 3 This is a schematic diagram of a water mist removal system for an unmanned driving sensing device according to some embodiments of this application;

[0045] Figure 4 These are schematic diagrams illustrating application scenarios based on some embodiments of this application. Detailed Implementation

[0046] The methods and systems provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0047] Example 1

[0048] like Figure 1 As shown, a method for removing water mist from an unmanned sensing device, used in underground mining environments, includes: S1, real-time acquisition of temperature and humidity data of the vehicle environment using at least two temperature and humidity sensors; wherein the temperature and humidity sensors are installed at the front of the vehicle to avoid the vehicle's own heat dissipation area; S2, filtering the acquired temperature and humidity data and calculating the historical temperature change rate. The process involves: S3, generating a feature vector containing current temperature, current humidity, and historical temperature change rate; S4, inputting the feature vector into a pre-trained lightweight neural network model for inference, outputting the current environmental water mist level, which includes four levels: safe, warning, light, and heavy; S5, determining the corresponding air supply temperature and fan speed parameters based on the water mist level using a preset mapping table, where the air supply temperature ranges from 0 to 40℃ and the fan speed ranges from 0 to 5000 r / min; S6, calculating the power control signal for the PTC heating element using a fuzzy PID algorithm based on the air supply temperature parameter, and simultaneously generating a fan speed control signal based on the fan speed parameter; and S7, controlling the PTC heating element and fan through an intelligent control unit, delivering the generated hot air through pipes to the base of the lidar and camera sensors, and then delivering the hot air to the sensor surface through the air outlet of the sensor base to remove water mist.

[0049] S1, Temperature and humidity sensor selection: (1) Measurement range: humidity range is 0%~100%RH, temperature range is -40~85℃. (2) Accuracy requirements: humidity ±5%RH, temperature ±0.5℃.

[0050] Temperature and humidity sensor installation: Sensor installation as follows Figure 2 As shown, it is recommended to install it at the front of the vehicle. In order to avoid the measured temperature and humidity being affected by the vehicle itself, three factors should be considered: (1) During the installation process, care should be taken to avoid the vehicle's own heat dissipation positions (such as the engine, air conditioning and other equipment); (2) Install it outside the vehicle and in direct contact with the environment; (3) Use at least two temperature and humidity sensors to avoid measurement errors from a single sensor.

[0051] Data transmission protocol: Data is transmitted to the intelligent computing unit via I2C / SPI interface; Modbus RTU or CAN FD protocol is used to ensure anti-interference in the vehicle environment.

[0052] Data Acquisition: The intelligent computing unit receives sensor data and converts it into double-precision temperature and humidity data according to relevant protocols. Simultaneously, the acquisition frequency must be set (e.g., once per second) to avoid high-frequency interference.

[0053] S2, Data Preprocessing: The Kalman filter algorithm is used to filter the temperature and humidity data. The process noise Q is set to a small value (0.01, 0.02) because the temperature and humidity in the downhole environment change relatively slowly. The measurement noise R is set to a large value (0.04, 0.08) to take into account the interference of downhole vibration and dust on the sensor. The noise parameter of humidity is greater than that of temperature because the accuracy of humidity sensor decreases in high humidity environment.

[0054] The filtered data was smoothed a second time using the moving average method. The window length was set to 10 sampling points, with a sampling frequency of 1Hz. The 10 points corresponded to a 10-second time window. The vibration and noise frequencies generated by downhole equipment (drilling rigs, loaders) are mainly in the range of 0.5 to 5Hz. The 10-second window can effectively filter out these high-frequency interferences while retaining the trend changes in temperature and humidity.

[0055] Saturation state detection is performed on the filtered humidity data. When the humidity value of 5 consecutive sampling points exceeds 95%RH, it is marked as humidity saturation state. Least square linear fitting is performed on the temperature data within the sliding window to obtain the historical temperature change rate. When humidity is saturated, the length of the sliding window is shortened from 10 sampling points to 5 sampling points to improve sensitivity to temperature gradient changes; the current temperature T, current humidity RH, and historical temperature change rate are used. The feature vector is obtained by combining the features. .

[0056] A lightweight neural network model based on TinyLSTM was constructed; the downhole environment was simulated in a temperature and humidity test chamber, and temperature and humidity data under condensation scenarios were collected as training set, with the temperature range set to -20℃ to 60℃ and the humidity range set to 10%RH to 100%RH.

[0057] Model training: The model is trained using input features and labels. During training, 80% of the dataset is used for training and 20% for validation.

[0058] Model evaluation: F1 score was used. Model evaluation is performed, and the result is defined as the harmonic mean of precision and recall. The formula for calculation is as follows: Where Recall is the recall rate and Precision is the precision rate, which are calculated from the confusion matrix and will not be explained in detail here.

[0059] Using feature vectors that are acquired and preprocessed in real time Input a pre-trained lightweight neural network model, and through forward propagation, output the probability distribution of four water mist levels. Select the level with the highest probability as the water mist level of the current environment.

[0060] S3, Domain Controller Selection: A domain controller capable of deploying deep learning (e.g., NVIDIA Jetson AGXOrin) will be used. Model Deployment: TensorRT will be used to lightweight the trained TinyLSTM model to accelerate the model inference process, resulting in a pre-trained lightweight neural network model. This includes: quantizing the 32-bit floating-point weights in the TinyLSTM model into 8-bit integers; calibrating the model on typical downhole temperature and humidity datasets to determine the dynamic range of activation values ​​for each layer, ensuring that the accuracy loss in water mist level judgment under high humidity saturation conditions is controlled within 2%, preventing missed detections of water mist condensation due to quantization errors; and merging the matrix operations of the input, forget, and output gates in the LSTM layers into a single large matrix operation, reducing the original requirement of four memory read / write operations. Time-series data processing is compressed to a single operation, reducing data transmission overhead by 75%, enabling the model to rapidly process continuous temperature and humidity data streams within the limited memory bandwidth of downhole vehicle-mounted equipment. To address the real-time requirements of downhole sensor data, the batch size is set to 1, avoiding data caching and ensuring that each temperature and humidity data sampling point can immediately enter the inference process, keeping the end-to-end latency from data acquisition to water mist level output within 50ms. Through data path optimization of the TensorRT inference engine, including tensor memory pre-allocation, inference graph optimization, and automatic kernel tuning, the data processing time for a single inference operation is compressed from 120ms to 30ms, ensuring smooth operation even with sudden temperature drops downhole. Under extreme conditions that cause rapid condensation of water mist, the system can complete the detection and initiate defogging measures within the golden response window (<100ms) at the initial stage of water mist formation.

[0061] The model is deployed using C++ code, and the algorithm's runtime must be less than 50ms. Model inference: During system runtime, this unit collects data in real time and inputs it into the model for inference. Output: The current environmental water mist level is obtained through model inference.

[0062] Air temperature and fan speed output: Based on the actual water mist removal effect, a table showing the relationship between water mist level, fan speed, and air temperature is provided, as shown in Table 1 below. Based on this table and the ambient water mist level, the supply air temperature and fan speed are retrieved and sent to the intelligent control unit.

[0063] Table 1 Comparison Table

[0064] Water mist level Wind speed requirement Supply air temperature (°C) Corresponding rotational speed (r / min) Safety Natural convection 0 0 Warning Low-speed air supply 20 1200 Mild Medium-speed air supply 30 3000 Severe High-speed air supply 40 5000

[0065] When the humidity value of 5 consecutive sampling points exceeds 95%RH, the air supply temperature corresponding to each water mist level in the mapping table will be increased by 2℃ and the fan speed will be increased by 10% to compensate for heat loss and water vapor retention caused by poor ventilation in the well.

[0066] When the water mist level is detected to jump directly from the safety level to light or heavy within a preset time, the heavy level demisting parameters are immediately executed, skipping the intermediate transition stage, in response to water mist condensation caused by a sudden drop in downhole temperature.

[0067] For mild and severe levels, a minimum continuous demisting time is set to prevent repeated condensation of water mist in the high-humidity environment downhole.

[0068] S4, Heating Control: Obtain the current actual temperature and air supply temperature Calculate temperature deviation And calculate the rate of change of deviation. Based on the water mist level output in step S3, set the control parameters for the fuzzy PID algorithm:

[0069] When the water mist level is safe, set ;

[0070] When water mist reaches the warning level, set ;

[0071] When the water mist level is light, set To respond quickly to defogging;

[0072] When the water mist level is severe, set The output power is limited to 100 to 150W;

[0073] Based on temperature deviation Deviation change rate Calculate the initial power of the PTC heating element based on the set PID control parameters;

[0074] The initial power of the PTC heating element is subjected to downhole explosion-proof safety constraints to generate the final output power. When the initial power of the PTC heating element is greater than 150W, set the output power. To ensure that the surface temperature of the PTC heating element does not exceed the explosion-proof safety threshold; when the output power is maintained at 150W and the temperature deviation is within 3 consecutive sampling cycles. When this is detected, it is considered an abnormal operating condition, and the output power is reduced. Reduce power to 100W and issue an alarm signal to prevent continuous high power output due to sensor malfunction: ,in, Target temperature With current supply air temperature The difference; These are the coefficients for proportional, integral, and derivative control, which need to be adjusted according to the specific application to achieve the best control effect.

[0075] Based on fan speed parameters and PTC heating element power This generates a synchronized fan speed control signal;

[0076] Example 2

[0077] like Figure 3 As shown, a water mist removal system for an autonomous driving sensing device based on a temperature and humidity sensor includes:

[0078] The data acquisition unit includes at least two temperature and humidity sensors, which are installed at the front of the vehicle to avoid the vehicle's own heat dissipation location, and are used to collect temperature and humidity data of the vehicle environment in real time.

[0079] The intelligent computing unit performs Kalman filtering and moving average processing on the collected temperature and humidity data, and calculates the historical temperature change rate. , forming feature vectors ;

[0080] Deploy a lightweight neural network model based on TinyLSTM, input feature vectors into the model for inference, and output water mist level;

[0081] The air supply temperature and fan speed parameters are determined using a mapping table based on the water mist level.

[0082] The power control signal of the PTC heating element is calculated using a fuzzy PID algorithm, and the fan speed control signal is generated.

[0083] The intelligent control unit includes a PTC heating element, which limits the maximum power of the PTC heating element to 150W to meet the explosion-proof requirements of the well; heating is performed according to the power control signal; and a fan generates airflow according to the speed control signal.

[0084] The cleaning unit includes an air supply duct, with its inlet end connected to the intelligent control unit, a PTC heating element, and a fan for delivering hot air.

[0085] like Figure 4 As shown, the intelligent control unit delivers hot air to the bases of sensors such as lidar and cameras through air ducts, and then delivers the hot air to the sensor surface through the air outlet of the sensor base to achieve the function of removing water mist from the sensor surface.

[0086] The foregoing illustrative description of the present application and its embodiments is not restrictive and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. The accompanying drawings are only one embodiment of the present application, and the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present application, such designs should fall within the scope of protection of this application. Furthermore, the word "comprising" does not exclude other elements or steps, and the word "a" preceding an element does not exclude the inclusion of "a plurality" of that element. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

Claims

1. A method for removing water mist from an unmanned sensing device, used in underground mining environments, characterized in that... include: S1 collects real-time temperature and humidity data of the vehicle environment through at least two temperature and humidity sensors; the temperature and humidity sensors are installed at the front of the vehicle to avoid the vehicle's own heat dissipation area. S2 filters the collected temperature and humidity data and calculates the historical temperature change rate. This forms a feature vector containing: current temperature, current humidity, and historical temperature change rate; S3 inputs the feature vector into a pre-trained lightweight neural network model, performs inference, and outputs the current water mist level. The water mist level includes four levels: safe, warning, light, and heavy. S4. Based on the water mist level, determine the corresponding air supply temperature parameters and fan speed parameters through a preset mapping table. The air supply temperature range is 0 to 40℃, and the fan speed range is 0 to 5000 r / min. S5: Based on the air supply temperature parameter, a fuzzy PID algorithm is used to calculate the power control signal for the PTC heating element; at the same time, based on the fan speed parameter, a fan speed control signal is generated. The S6 controls the PTC heating element and fan through the intelligent control unit. The generated hot air is delivered to the base of the lidar and camera sensor through the pipe, and then delivered to the sensor surface through the air outlet of the sensor base to remove water mist. Specifically, S4 determines the corresponding air supply temperature parameters and fan speed parameters based on the water mist level and a preset mapping table, including: When the humidity value of 5 consecutive sampling points exceeds 95%RH, the air supply temperature corresponding to each water mist level in the mapping table will be increased by 2℃ and the fan speed will be increased by 10% to compensate for heat loss and water vapor retention caused by poor ventilation in the well. When the water mist level is detected to jump directly from the safety level to the light or heavy level within a preset time, the heavy level demisting parameters are immediately executed, skipping the intermediate transition stage, in response to water mist condensation caused by a sudden drop in downhole temperature. For mild and severe levels, a minimum continuous defogging time is set to prevent repeated condensation of water mist in the high-humidity environment downhole.

2. The water mist removal method for the unmanned sensing device according to claim 1, characterized in that: Temperature data collection range: -40℃ to 85℃; Humidity data collection range: 0 to 100%RH; The sampling frequency is once per second.

3. The water mist removal method for the unmanned sensing device according to claim 2, characterized in that: S2 filters the collected temperature and humidity data and calculates the historical temperature change rate. This forms a feature vector containing: current temperature, current humidity, and historical temperature change rate, including: The Kalman filter algorithm is used to filter the temperature and humidity data; The filtered data was smoothed a second time using the moving average method, with the window length set to 10 sampling points to eliminate high-frequency noise generated by downhole equipment. The filtered humidity data is subjected to saturation detection. When the humidity value of 5 consecutive sampling points exceeds 95%RH, it is marked as humidity saturation. The historical temperature change rate is obtained by performing a least-squares linear fit on the temperature data within the sliding window. When the humidity is saturated, the length of the sliding window is shortened from 10 sampling points to 5 sampling points to improve the sensitivity to temperature gradient changes. The current temperature T, current humidity RH, and historical temperature change rate are used. The feature vector is obtained by combining the features. .

4. The water mist removal method for the unmanned sensing device according to claim 2, characterized in that: S3 inputs the feature vectors into a pre-trained lightweight neural network model for inference and outputs the current environmental water mist level. The water mist level includes four levels: safe, warning, light, and heavy. Construct a lightweight neural network model based on TinyLSTM; The underground environment was simulated in a temperature and humidity test chamber, and temperature and humidity data under condensation scenarios were collected as a training set. The temperature range was set to -20℃ to 60℃ and the humidity range was set to 10%RH to 100%RH. Based on the training set, the model is trained using the cross-entropy loss function and the Adam optimizer, and the F1 score is used as the model evaluation metric to obtain the trained TinyLSTM model. TensorRT is used to lightweight the trained TinyLSTM model to accelerate the model inference process and obtain a pre-trained lightweight neural network model. Using feature vectors that are acquired and preprocessed in real time Input a pre-trained lightweight neural network model, and through forward propagation, output the probability distribution of four water mist levels. Select the level with the highest probability as the water mist level of the current environment.

5. The water mist removal method for the unmanned sensing device according to claim 4, characterized in that: Constructing a lightweight neural network model based on TinyLSTM, including: The input layer is set to a dimension of 3 to receive feature vectors. ; The LSTM hidden layer node number is set to 32 and the time window length is 10 to obtain the rapid condensation process caused by small temperature changes under high humidity and saturation conditions in the downhole environment. The output layer is set to have 4 nodes, corresponding to four water mist levels: safety, warning, light, and heavy. The heavy level is used to identify extreme condensation conditions where the humidity exceeds 95% and the temperature drops rapidly.

6. The water mist removal method for an unmanned sensing device according to any one of claims 2 to 4, characterized in that: S4, the mapping table, includes: When the water mist level is safe, set the air supply temperature to 0℃ and the fan speed to 0r / min. When the water mist level is at the warning level, set the air supply temperature to 20℃ and the fan speed to 1200r / min to prevent water mist from condensing on the sensor surface. When the water mist level is light, set the air supply temperature to 30℃ and the fan speed to 3000r / min to remove the formed water mist; When the water mist level is severe, set the air supply temperature to 40℃ and the fan speed to 5000r / min to prevent water mist condensation.

7. The water mist removal method for the unmanned sensing device according to claim 1, characterized in that: S5, based on the supply air temperature parameter, uses a fuzzy PID algorithm to calculate the power control signal for the PTC heating element; simultaneously, based on the fan speed parameter, it generates a fan speed control signal, including: Get the current actual temperature and air supply temperature Calculate temperature deviation And calculate the rate of change of deviation. ; Based on the water mist level output in step S3, set the control parameters for the fuzzy PID algorithm: When the water mist level is safe, set ; When water mist reaches the warning level, set ; When the water mist level is light, set To respond quickly to defogging; When the water mist level is severe, set The output power is limited to 100 to 150W; Based on temperature deviation Deviation change rate Calculate the initial power of the PTC heating element based on the set PID control parameters; The initial power of the PTC heating element is subjected to downhole explosion-proof safety constraints to generate the final output power. ; Based on fan speed parameters and PTC heating element power This generates a synchronized fan speed control signal; in, These are the coefficients for proportional, integral, and derivative control, respectively.

8. The water mist removal method for the unmanned sensing device according to claim 7, characterized in that: The initial power of the PTC heating element is subjected to downhole explosion-proof safety constraints to generate the final output power. ,include: When the initial power of the PTC heating element is greater than 150W, set the output power. To ensure that the surface temperature of the PTC heating element does not exceed the explosion-proof safety threshold; When the output power is detected to remain at 150W for three consecutive sampling cycles and the temperature deviation is... When this is detected, it is considered an abnormal operating condition, and the output power is reduced. Reduce the power to 100W and issue an alarm signal to prevent continuous high power output due to sensor failure.

9. A water mist removal system for an unmanned sensing device, used in underground mining environments, characterized in that, include: The data acquisition unit includes at least two temperature and humidity sensors, which are installed at the front of the vehicle to avoid the vehicle's own heat dissipation location, and are used to collect temperature and humidity data of the vehicle environment in real time. The intelligent computing unit performs Kalman filtering and moving average processing on the collected temperature and humidity data, and calculates the historical temperature change rate. , forming feature vectors ; Deploy a lightweight neural network model based on TinyLSTM, input feature vectors into the model for inference, and output water mist level; The air supply temperature and fan speed parameters are determined using a mapping table based on the water mist level. The power control signal of the PTC heating element is calculated using a fuzzy PID algorithm, and the fan speed control signal is generated. The intelligent control unit includes a PTC heating element, which limits the maximum power of the PTC heating element to 150W to meet the explosion-proof requirements of the well; heating is performed according to the power control signal; and a fan generates airflow according to the speed control signal. The cleaning unit includes an air supply duct, with its inlet end connected to the intelligent control unit, PTC heating element, and fan for delivering hot air; The air supply temperature and fan speed parameters are determined based on the water mist level using a mapping table, including: When the humidity value of 5 consecutive sampling points exceeds 95%RH, the air supply temperature corresponding to each water mist level in the mapping table will be increased by 2℃ and the fan speed will be increased by 10% to compensate for heat loss and water vapor retention caused by poor ventilation in the well. When the water mist level is detected to jump directly from the safety level to the light or heavy level within a preset time, the heavy level demisting parameters are immediately executed, skipping the intermediate transition stage, in response to water mist condensation caused by a sudden drop in downhole temperature. For mild and severe levels, a minimum continuous defogging time is set to prevent repeated condensation of water mist in the high-humidity environment downhole.

Images