A dust removal air pipe unblocking robot automatic positioning system

Abstract

This invention relates to the field of robot automatic positioning technology, specifically an automatic positioning system for a dust removal and duct clearing robot. The system includes: a data acquisition module for acquiring the robot's inertial parameters, hydraulic environment parameters, a preset digital map of the pipeline, a preset dead reckoning high-risk threshold, and a preset dead reckoning failure threshold; a state estimation module for determining the current estimated position and estimated velocity, and also for determining the hydraulic feedback feature vector; a risk assessment module for calculating dead reckoning and physical interference factors, and also for determining the predicted future cumulative risk and total risk value; and a closed-loop correction module for generating graded correction instructions and updating the estimated position to a reliable anchor point. This invention solves the problem of irrecoverable positioning loss, ensuring robot safety.

Description

Technical Field

[0001] This invention relates to the field of robot automatic positioning technology, specifically to an automatic positioning system for a dust removal duct unblocking robot. Background Technology

[0002] In industries such as metallurgy and power, unclogging dust removal ducts is crucial for maintaining system operation. Unclogging robots need to go deep into narrow, winding blind duct environments without external signal coverage to perform tasks. In such environments, the robot's high-precision autonomous positioning is the basic prerequisite for achieving efficient operation and safe return.

[0003] Traditional positioning schemes heavily rely on inertial navigation technology. This technology uses inertial measurement units for dead reckoning, but in blind tubes without external calibration, its inherent cumulative drift problem cannot be suppressed. As the operating distance increases, positioning errors inevitably accumulate, eventually leading to complete positioning failure and causing the robot to become lost. More seriously, water flow operations themselves generate severe physical interference, which not only accelerates the drift of inertial sensors but also pollutes environmental sensing signals. Existing technologies generally lack effective utilization of environmental information and cannot integrate hydraulic, acoustic, and other signals to assist positioning. At the same time, after the system detects positioning drift, it lacks a forward-looking risk assessment and an effective closed-loop correction mechanism. Once positioning fails, it cannot be recovered, posing a significant safety hazard. Therefore, how to provide a highly robust positioning system that can integrate multi-source sensing, assess risks in real time, and has the ability to recover from failure is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides an automatic positioning system for a dust removal duct unblocking robot. Specifically, the technical solution of the present invention includes:

[0005] The data acquisition module is used to acquire the robot's inertial parameters, hydraulic environment parameters, preset pipeline digital map, preset dead reckoning high-risk threshold and preset dead reckoning failure threshold. The inertial parameters include three-axis acceleration vector and three-axis angular velocity vector, and the hydraulic environment parameters include water pressure and water flow impact acoustic signal.

[0006] The state estimation module is used to determine the current estimated position and estimated velocity based on the three-axis acceleration vector and the three-axis angular velocity vector through a preset dead reckoning model; the state estimation module is also used to determine the hydraulic feedback feature vector based on the water pressure and water impact acoustic signal through a preset feature extraction network.

[0007] The risk assessment module is used to comprehensively estimate the location, hydraulic feedback feature vector, and pipeline digital map to calculate dead entropy; the risk assessment module is also used to calculate physical disturbance factor based on water flow pressure; the risk assessment module is also used to determine the predicted future cumulative risk based on physical disturbance factor, dead entropy, and estimated speed through a preset entropy increase prediction model; the risk assessment module is also used to determine the total risk value by combining dead entropy and predicted future cumulative risk.

[0008] The closed-loop correction module is used to generate graded correction instructions based on the total risk value, dead reckoning, preset dead reckoning high-risk threshold and preset dead reckoning failure threshold; the closed-loop correction module is also used to update the estimated position to a reliable anchor point when the dead reckoning is less than or equal to the preset dead reckoning high-risk threshold.

[0009] Specifically, when the total risk value is greater than the preset dead reckoning high-risk threshold but less than or equal to the preset dead reckoning failure threshold, a first-level correction is executed to reduce the robot's forward command speed.

[0010] Specifically, when the total risk value exceeds the preset dead reckoning failure threshold, a secondary correction is performed, the forward command speed is set to zero, the retraction mechanism is activated to retract the command, and when the robot retreats to the vicinity of the trusted anchor point, the estimated position is reset to the trusted anchor point.

[0011] Optionally, the state estimation module determines the current estimated position and estimated velocity based on the triaxial acceleration vector and triaxial angular velocity vector, including:

[0012] Based on the three-axis acceleration vector and the three-axis angular velocity vector, a basic kinematic model for dead reckoning is constructed.

[0013] By using the basic kinematic model and the state at the previous moment, the estimated position and estimated velocity at the current moment are recursively estimated.

[0014] Optionally, the state estimation module determines the hydraulic feedback feature vector based on the water flow pressure and the water flow impact acoustic signal, including:

[0015] The water flow pressure and water flow impact acoustic signals are used as time-series inputs;

[0016] The time-series signal is processed by a pre-defined feature extraction network and converted into a structured hydraulic feedback feature vector.

[0017] Optionally, the risk assessment module calculates dead entropy, including:

[0018] Call upon the estimated location, hydraulic feedback feature vector, and digital map of the pipeline;

[0019] By analyzing the estimated location on the pipeline digital map output by the probabilistic model, the conditional probability of the hydraulic feedback feature vector was observed.

[0020] Detachment entropy is determined by calculating the negative log-likelihood of the conditional probability.

[0021] Optionally, the risk assessment module calculates the physical interference factor, including:

[0022] Real-time monitoring of water flow pressure time series;

[0023] Calculate the instantaneous rate of change of water pressure;

[0024] The physical interference factor is quantified based on the instantaneous rate of change of water flow pressure and the recent average pressure, combined with a preset sensitivity weighting coefficient.

[0025] Optionally, the risk assessment module determines the predicted future cumulative risk, including:

[0026] Call upon the physical interference factor, dead reckoning, and estimated speed;

[0027] The physical interference factor, dead entropy, and estimated speed are input into a pre-defined entropy increase prediction model to determine the predicted future cumulative risk.

[0028] Optionally, the risk assessment module determines the total risk value, including:

[0029] Preset weighting coefficients are assigned to dead reckoning and predicted future cumulative risk;

[0030] The total risk value is determined by weighted summation of dead reckoning and predicted future cumulative risk.

[0031] Optionally, the first-level correction includes:

[0032] The total risk value is determined to be greater than the preset dead reckoning high-risk threshold and less than or equal to the preset dead reckoning failure threshold.

[0033] Actively reduce the robot's forward speed and keep the winding mechanism in a forward state in an attempt to enable the system to self-correct.

[0034] Optionally, the secondary correction includes:

[0035] The total risk value is determined to be greater than the preset dead reckoning failure threshold;

[0036] Immediately set the forward speed to zero and send a retraction command to the tube winding mechanism;

[0037] When the robot passively retreats to the vicinity of the trusted anchor point, it stops retraction, forcibly resets the positioning system state to the trusted anchor point, and resets the dead reckoning to zero.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. This system innovatively integrates inertial parameters and hydraulic environment parameters, converts water flow pressure and acoustic signals into usable positioning features, and compares them with the pipeline digital map, overcoming the defect of traditional inertial navigation inevitably accumulating drift in blind pipe environments;

[0040] 2. This system quantifies the degree of discrepancy between inertial calculation and environmental perception in real time by introducing dead reckoning, i.e., positioning disorientation degree; at the same time, it quantifies physical interference sources such as water pressure fluctuations by calculating physical interference factors, and realizes a comprehensive assessment of the causes and current status of risks.

[0041] 3. This system has constructed an entropy increase prediction model, which integrates the current degree of disorientation, physical interference causes, and robot speed. It can proactively predict the risk of positioning failure that will occur in the future, enabling the system to intervene in advance rather than responding with a lag, thus improving proactive safety.

[0042] 4. This system has a hierarchical closed-loop correction capability: it actively slows down in high-risk situations to attempt self-correction; in case of failure, it executes a highly resilient strategy, namely, stopping operation, reverting to the last trusted anchor point and resetting the positioning state, which solves the pain point of being unable to recover after positioning loss and ensures robot safety. Attached Figure Description

[0043] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0044] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0046] Example 1:

[0047] Please see Figure 1 An automatic positioning system for a dust removal duct unblocking robot, comprising:

[0048] The data acquisition module is used to acquire the robot's inertial parameters, hydraulic environment parameters, preset digital pipeline map, preset dead reckoning high-risk threshold, and preset dead reckoning failure threshold. The inertial parameters include three-axis acceleration vectors and three-axis angular velocity vectors, and the hydraulic environment parameters include water pressure and water flow impact acoustic signals.

[0049] The state estimation module is used to determine the current estimated position and estimated velocity based on the three-axis acceleration vector and the three-axis angular velocity vector through a preset dead reckoning model; the state estimation module is also used to determine the hydraulic feedback feature vector based on the water pressure and water impact acoustic signals through a preset feature extraction network;

[0050] The risk assessment module is used to comprehensively estimate location, hydraulic feedback feature vectors, and pipeline digital maps to calculate dead reckoning; it is also used to calculate physical disturbance factors based on water flow pressure; furthermore, it is used to determine the predicted future cumulative risk based on the physical disturbance factors, dead reckoning, and estimated speed using a pre-set entropy increase prediction model; and finally, it is used to determine the total risk value by combining dead reckoning and the predicted future cumulative risk.

[0051] The closed-loop correction module is used to generate graded correction instructions based on the total risk value, dead reckoning, a preset dead reckoning high-risk threshold, and a preset dead reckoning failure threshold. The closed-loop correction module is also used to update the estimated position to a reliable anchor point when the dead reckoning is less than or equal to the preset dead reckoning high-risk threshold.

[0052] Specifically, when the total risk value is greater than the preset dead reckoning high-risk threshold but less than or equal to the preset dead reckoning failure threshold, a first-level correction is executed, reducing the robot's forward command speed;

[0053] Specifically, when the total risk value exceeds the preset dead reckoning failure threshold, a secondary correction is performed, the forward command speed is set to zero, the retraction mechanism is activated to retract the command, and when the robot retreats to the vicinity of the trusted anchor point, the estimated position is reset to the trusted anchor point.

[0054] This embodiment provides an automatic positioning system for a dust removal duct unblocking robot. The system is a complete and self-consistent technical closed loop, designed to achieve highly robust positioning of the robot in blind duct environments. The system includes four core modules: a data acquisition module, a state estimation module, a risk assessment module, and a closed-loop correction module.

[0055] The data acquisition module aims to provide the necessary raw data input for all subsequent estimations and evaluations; in this embodiment, it is implemented in the following way:

[0056] Acquiring inertial parameters: The robot's inertial parameters, namely the three-axis acceleration vectors, are acquired in real time through an inertial measurement unit (IMU) installed on the robot body. With the three-axis angular velocity vector This data stream forms the basis for subsequent dead reckoning.

[0057] Acquiring hydraulic environment parameters: Hydraulic environment parameters, namely water flow pressure, are collected in real time using pressure sensors and underwater audio devices installed near the robot's nozzle. and water flow impact acoustic signals This data stream serves as the basis for environmental perception.

[0058] Obtain preset parameters: During initialization, the module loads a preset pipeline digital map from memory. Preset dead reckoning high-risk threshold and the preset dead reckoning failure threshold ;

[0059] Pipeline Digital Map It refers to a priori map generated based on engineering blueprints, which stores the expected geometric features of pipeline locations such as straight pipes and bends. Its purpose is to provide a baseline truth for the calculation of dead reckoning.

[0060] High Threshold of Dead Entropy and dead reckoning failure threshold These refer to two dimensionless scalars, which serve as the basis for the risk assessment module to make graded judgments. Their values ​​can be obtained by reproducing the positioning drift and failure process on the test platform and statistically calibrating the changes in dead reckoning H.

[0061] The state estimation module aims to calculate the robot's current location and environmental conditions in real time. In this embodiment, it performs two tasks in parallel:

[0062] Dead reckoning: This module is based on three-axis acceleration vectors. With the three-axis angular velocity vector The estimated current position is determined by a pre-set dead reckoning model, i.e., a classical basic kinematic model. With estimated speed This process is achieved through recursive estimation, providing a basic position solution in the absence of external signals.

[0063] Feature extraction: This module is also based on water flow pressure. Acoustic signals of water flow impact Through a pre-set feature extraction network Determine the hydraulic feedback eigenvector ;

[0064] Hydraulic feedback feature vector It refers to a structured, dimensionless vector whose function is to convert the high-noise raw P,S signal into features that the system can understand and that encode environmental events such as passing through bends and encountering traffic jams.

[0065] The risk assessment module aims to quantify the current and future failure risks of the positioning system; in this embodiment, it performs three key calculations:

[0066] Calculate dead entropy H: This module comprehensively estimates the position. Hydraulic feedback eigenvectors and pipeline digital maps Calculate dead entropy ;

[0067] deadlock This refers to an indicator that quantifies the degree of disorientation in location tracking; it is calculated using a probability model. Assessment on map of Location observed The probability of a feature, and its negative log-likelihood value. When the IMU calculates With environmental perception In the map When there is a serious contradiction under the support of [the relevant entity], the probability is low and the H value is high.

[0068] Calculate the physical interference factor This module is based on water flow pressure. Calculate the physical interference factor ;

[0069] Physical interference factors This refers to an index that quantifies the physical causes of location failure due to drastic fluctuations in water pressure; its calculation method is as follows: ,in It is the instantaneous rate of change of pressure. This is the recent average pressure. It is a sensitivity weight;

[0070] Calculate and predict risk Total risk R:

[0071] This module is based on physical interference factors. deadlock and estimated speed Through a pre-set entropy increase prediction model Determine the predicted future cumulative risk Forecasted future cumulative risks It refers to the predicted value of the future entropy accumulation that will result from the system continuing to move forward in its current state;

[0072] This module incorporates dead reckoning. Accumulated risks in the predicted future By weighted summation Determine the total risk value ;

[0073] The purpose of the closed-loop correction module is to perform resilience-prioritized closed-loop control of the robot's positioning state and physical motion based on the assessed risks.

[0074] Generate tiered instructions: This module is based on the total risk value. deadlock Preset dead reckoning high-risk threshold Compared with the preset dead reckoning failure threshold Generate hierarchical correction instructions;

[0075] Update Trusted Anchor Point: This module has a trusted anchor point register. Trusted anchors This refers to the dead reckoning when the system's positioning is reliable. Less than or equal to the preset dead entropy high-risk threshold The last high-confidence estimated location stored in real time When the system enters a high-risk or failure state, Stop updating and retain its last safe value;

[0076] Execute Level 1 Correction: Where, when the total risk value The dead reckoning is higher than the preset high-risk threshold. And less than or equal to the preset dead reckoning failure threshold At that time, a first-level correction is executed, reducing the robot's forward speed. Meanwhile, the tube winding mechanism is kept in a forward position; this is intended to attempt self-correction by slowing down, causing the H value to drop.

[0077] Execute a second-level correction: where, when the total risk value The failure threshold is greater than the preset dead reckoning threshold. At that time, a second-level correction is executed; this correction follows a high-cost resilience-first strategy.

[0078] Set the forward command speed to zero. ;

[0079] Activate the retraction command for the tube winding mechanism ;

[0080] And the robot reverts to the trusted anchor point Stop retrieving when the area is nearby;

[0081] Reset the estimated location to a trusted anchor point and reset dead reckoning ;

[0082] This embodiment constructs a system through the coordinated operation of the four modules mentioned above, from inertial sensing to hydraulic environment sensing, and then to the current risk H and future risks. The system integrates evaluation and hierarchical closed-loop correction to form a complete positioning system. It solves the problem that traditional inertial navigation inevitably fails and cannot be recovered in blind pipes. By introducing hydraulic feedback, dead entropy and forward-looking risk prediction, and combining a two-level correction strategy of resetting to the anchor point, it provides the blockage clearing robot with highly robust and resilient autonomous positioning capabilities in extremely harsh environments.

[0083] Example 2:

[0084] The state estimation module determines the current estimated position and estimated velocity based on the triaxial acceleration vector and triaxial angular velocity vector, including:

[0085] Based on the three-axis acceleration vector and the three-axis angular velocity vector, a basic kinematic model for dead reckoning is constructed;

[0086] By using the basic kinematic model and the state at the previous moment, the estimated position and estimated velocity at the current moment are recursively estimated.

[0087] This embodiment specifically illustrates how the state estimation module determines the estimated position and estimated velocity based on data;

[0088] The state estimation module is based on a triaxial acceleration vector. With the three-axis angular velocity vector Determine the current estimated location With estimated speed Its specific implementation includes:

[0089] Based on triaxial acceleration vector With the three-axis angular velocity vector We construct a basic kinematic model for dead reckoning; this model is derived from the dead reckoning equations in classical inertial navigation.

[0090] Using the basic kinematic model and the state at the previous moment, the estimated position and estimated velocity at the current moment are recursively estimated.

[0091] In this embodiment, the basic kinematic model is defined by the following recursive formula:

[0092] ;

[0093] ;

[0094] The purpose of this model is to provide basic, continuous position estimation using IMU data in blind-scan environments without external positioning signals. This basic kinematic model employs the Euler method for discrete-time integration, a simplified implementation used to ensure real-time computational efficiency. In other embodiments, higher-order integrators such as the Runge-Kutta method can be used to achieve more efficient real-time computation. Achieve higher dead reckoning fidelity when the motion is large or changes drastically;

[0095] in: The estimated position and velocity vectors at the current moment are the output of this step;

[0096] For the previous moment The estimated position vector; For the previous moment The estimated velocity vector, together with the estimated velocity vector, constitutes the recursive input state for this step;

[0097] The acceleration measured by the IMU at time t-1 is derived from the data acquisition module;

[0098] The attitude matrix at time t-1 is derived from the... The data is obtained by integration from the data acquisition module.

[0099] This is the gravity vector, which originates from prior calibration;

[0100] The time step is derived from the system clock;

[0101] The system iterates through the above formula, utilizing the previous time step... And the currently collected Autonomously calculates the three-dimensional position within the pipeline and speed ;

[0102] This embodiment ensures the system possesses the fundamental capability for autonomous positioning in blind-sight environments by clearly defining the basic kinematic model for dead reckoning and its recursive estimation process; it uses the three-axis acceleration vector representing linear motion characteristics from the IMU's raw data. With the three-axis angular velocity vector characterizing attitude change features The extracted and deeply fused information was transformed into structured state information necessary for subsequent dead entropy assessment and entropy increase prediction. .

[0103] Example 3:

[0104] The state estimation module determines the hydraulic feedback feature vector based on water pressure and water impact acoustic signals, including:

[0105] Water pressure and water impact acoustic signals are used as time-series inputs;

[0106] The time-series signal is processed by a pre-defined feature extraction network and converted into a structured hydraulic feedback feature vector.

[0107] This embodiment specifically illustrates how the state estimation module uses environmental information to determine the hydraulic feedback feature vector;

[0108] The state estimation module is based on water flow pressure. Acoustic signals of water flow impact Determine the hydraulic feedback feature vector Its specific implementation includes:

[0109] Water flow pressure Acoustic signals of water flow impact As a timing signal input;

[0110] Through a pre-set feature extraction network Process the time-series signals and convert them into structured hydraulic feedback feature vectors. In this embodiment, the conversion process is implemented using a deep learning model:

[0111] ;

[0112] To further clarify the network structure, in this embodiment, This can be specifically implemented as a two-channel convolutional neural network: and The temporal signal is fed into two parallel one-dimensional convolutional branches; each branch contains two convolutional layers, for example, with a kernel size of 3, a stride of 1, ReLU activation, and a max-pooling layer; then the output features of the two branches are flattened and concatenated; finally, a fully connected layer, for example, with 128 neurons, is passed through, and ReLU activation outputs a fixed-dimensional hydraulic feedback feature vector. ;

[0113] The purpose of this model is to address high-noise, non-landmark environments. The timing signal is converted into structured features that the positioning system can understand and use to correct IMU drift. IMU data itself cannot perceive the environment, but the interaction between water flow and pipes contains rich information about the environmental structure.

[0114] in: The hydraulic feedback feature vector is a structured, dimensionless vector that encodes environmental events such as passing through bends and encountering blockages; it is one of the core inputs for subsequent dead entropy calculations.

[0115] The feature extraction network is a pre-defined deep learning model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN), which is used to process temporal data.

[0116] The input raw pressure and acoustic timing signals originate from the data acquisition module;

[0117] These are network weight parameters;

[0118] parameter Method for determining: parameters The topology is determined through offline training in experimental pipelines with known topologies; deep reinforcement learning (DRL) or self-supervised learning methods can be used for training on a large amount of experimental data. To make its output It exhibits the greatest correlation with known pipeline landmarks such as bends and junctions;

[0119] This embodiment introduces a preset feature extraction network. It innovatively solves the fundamental flaw of inertial navigation's inability to perceive the environment; it utilizes features that the system cannot directly leverage. The hydraulic-acoustic signal was converted into a hydraulic feedback feature vector that can be used to correct drift. This provides a key technical bridge for achieving coupled modeling of inertial inference and environmental perception.

[0120] Example 4:

[0121] The risk assessment module calculates dead entropy, including:

[0122] Access estimated location, hydraulic feedback feature vector, and digital map of the pipeline;

[0123] By analyzing the estimated location on the digital map of the pipeline output by the probabilistic model, the conditional probability of the hydraulic feedback feature vector was observed.

[0124] Detachment entropy is determined by calculating the negative log-likelihood of the conditional probability.

[0125] This embodiment specifically illustrates how the risk assessment module calculates the core indicator of the degree of disorientation, namely dead reckoning;

[0126] The risk assessment module calculates dead entropy. Its specific implementation includes:

[0127] Call the estimated location Hydraulic feedback feature vector and pipeline digital maps ;

[0128] Output from the probabilistic model in the pipeline digital map Estimated position on The hydraulic feedback eigenvector was observed. conditional probability ;

[0129] Detachment entropy is determined by calculating the negative log-likelihood of the conditional probability. In this embodiment, the calculation formula is:

[0130] ;

[0131] To further clarify this probability model, in this embodiment, the model... This can be achieved using a small neural network: from pipeline digital maps Search Map features corresponding to the location ;Will Extracted map features Observation features obtained from the state estimation module Perform concatenation; input the concatenated vector into a classifier containing two fully connected layers and a sigmoid activation output layer. The 0-1 scalar values ​​of this output layer represent the conditional probabilities. ;

[0132] The purpose of this model is to quantify the reliability of a positioning system, i.e., the degree of disorientation; its inertial estimation... It will accumulate drift, while environmental perception Reflects the real physical environment; dead reckoning The design motivation is to detect the degree of contradiction between these two in real time;

[0133] in: H is dead entropy, a dimensionless scalar value used as a quantitative indicator of the reliability of a positioning system; a low H value indicates... and In the map With the support of [unclear], the positioning is highly consistent and reliable; a surge in the H value indicates a serious contradiction between the two, and the system has experienced significant drift or loss of direction.

[0134] Conditional probability is the probability value output by a probability model; it is used in pipeline digital maps. Defined Location, observed hydraulic feedback feature vector How likely is it?

[0135] When the system is working well It is accurate. The environmental characteristics reflected in the map exist The location records are consistent, at this time High probability Low value; when drift causes Misalignment, and When it is still perceived as a straight pipe, The probability is extremely low. The value spiked; to further clarify this probability model, in this embodiment, It can be built through offline learning, for example, by collecting a large number of samples in an experimental pipeline. The data pairs are used to train a probability density estimation network or Gaussian mixture model to fit the conditional probability distribution. When the map... exist Expected characteristics of location and When the features represented by the vectors do not match, the model It will then output a very low probability value.

[0136] This embodiment introduces dead reckoning. This innovative concept elevates the positioning problem from simple location calculation to the level of reliability assessment; it uses a probabilistic model to transform the inertial estimation results... Characteristics of hydraulic environment perception Information theory integration was implemented, providing a real-time, quantitative indicator to assess the current degree of disorientation of the positioning system. This is the core decision-making tool for all subsequent risk warnings and graded corrections.

[0137] Example 5:

[0138] The risk assessment module calculates the physical interference factor, including:

[0139] Real-time monitoring of water flow pressure time series;

[0140] Calculate the instantaneous rate of change of water pressure;

[0141] The physical interference factor is quantified based on the instantaneous rate of change of water flow pressure and the recent average pressure, combined with a preset sensitivity weighting coefficient.

[0142] This embodiment specifically illustrates how the risk assessment module calculates the index for locating physical interference sources, namely the physical interference factor;

[0143] The risk assessment module calculates the physical interference factor. Its specific implementation includes:

[0144] Real-time monitoring of water flow pressure Time series;

[0145] Calculate water flow pressure instantaneous rate of change ;

[0146] Instantaneous rate of change of water flow pressure Compared with recent average pressure And combined with preset sensitivity weighting coefficients Quantitatively determine physical interference factors In this embodiment, the quantification formula is:

[0147] ;

[0148] The purpose of this model is to quantify the severity of the physical layer disturbance caused by water pressure fluctuations to the positioning base; dead reckoning. The assessment focuses on the contradictions in the results, while physical interference factors... The aim is to assess the cause; severe fluctuations in water pressure can simultaneously destroy two fundamental aspects of positioning at the physical level: compromising the accuracy of acceleration inputs for dead reckoning; and weakening hydraulic feedback signals, leading to issues in feature extraction. The ambiguity is that this embodiment simplifies and characterizes physical disturbances as a normalized rate of pressure change, which is the most significant physical cause of positioning failure in practice. In other embodiments, more complex physical quantities such as the spectral characteristics of pressure and vibration frequency can also be introduced to construct the ambiguity. ;

[0149] in: The physical interference factor is a dimensionless scalar that proactively indicates whether the two data bases for positioning are being contaminated by physical interference.

[0150] The instantaneous rate of change of pressure originates from the... Real-time differentiation calculation of time series;

[0151] The recent average pressure stems from the... Smoothing calculation of time series data, used as a baseline;

[0152] This is the sensitivity weighting coefficient, with the dimension of time. to make Dimensionless; It is a preset, extremely small positive number, for example ,Should Already As part of the denominator in the calculation formula, it ensures the robustness of the model;

[0153] To further clarify the parameters The method of determination The calibration was determined through experiments; specifically, pressure drops of different amplitudes and rates were artificially introduced onto the test platform, and the corresponding calibration pressure signals were collected. and positioning accuracy ; By using regression analysis methods such as least squares, the calibration is performed. Value, making based on Calculated interference factor The negative correlation with positioning accuracy is the most significant;

[0154] This embodiment designs physical interference factors. This expands the risk assessment module's capabilities from perceiving the current situation (H) to understanding the underlying causes. It provides the system with a key physical layer indicator for quantifying the direct physical causes of failure, which is the core data input for achieving the next stage of entropy increase prediction.

[0155] Example 6:

[0156] The risk assessment module identifies the predicted cumulative risks in the future, including:

[0157] Invoke the physical interference factor, dead reckoning, and estimated speed;

[0158] The physical interference factor, dead entropy, and estimated speed are input into a pre-defined entropy increase prediction model to determine the predicted future cumulative risk.

[0159] This embodiment illustrates in detail how the risk assessment module proactively determines future risks;

[0160] The risk assessment module determines the predicted future cumulative risk. Its specific implementation includes:

[0161] Call physical interference factor deadlock and estimated speed ;

[0162] Physical interference factors deadlock and estimated speed Input into the preset entropy increase prediction model To determine the predicted cumulative risk in the future In this embodiment, the prediction model is:

[0163] ;

[0164] To further clarify the prediction network, in this embodiment, This can be specifically implemented as a feedforward neural network: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] scalar And the scalar magnitude of the estimated velocity The input vector is concatenated into a three-dimensional input vector; this vector then passes through two fully connected hidden layers, for example, one with 64 neurons and the other with 32 neurons, both using the ReLU activation function; finally, it passes through a single-neuron linear output layer, outputting a single scalar value. ;

[0165] The purpose of this model is to predict the total entropy, or risk, that a system will accumulate in the future if it continues its current state; the system must not only assess the current entropy. Furthermore, it is necessary to predict future entropy; this invention argues that future risks... It is the result of the combined effect of three factors: As a fundamental risk; As a risk factor, when High time, The probability of a future surge is high; As a risk amplifier, when At higher altitudes, drift accumulates faster;

[0166] in: The cumulative risk for the predicted future is a dimensionless scalar, whose role is to provide a forward-looking risk assessment for the system, which can be used to trigger early closed-loop corrections.

[0167] An entropy increase prediction model is a pre-defined prediction network, such as the value network in a deep reinforcement learning (DRL) framework, whose function is to fit the target value. Status and Future Risks The complex relationship between them;

[0168] Entropy increase prediction model The network parameter set, which covers all trainable connection weight matrices and bias vectors in the network;

[0169] Network parameter set Method for determining: network parameter set Can be used with feature extraction networks parameters They were trained together under the DRL framework; the training objective was to make Able to accurately predict in specific In this state, the final entropy accumulation results from continuing the action forward;

[0170] This embodiment introduces an entropy increase prediction model. This invention achieves a forward-looking risk assessment; it transforms risk assessment from passively responding to current disorientation (H) to actively predicting future disorientation. This forward-looking approach allows closed-loop corrections to be implemented even before the failure threshold is reached in this phase (H). At that time, because The system intervenes early when the temperature rises, greatly enhancing the resilience of the positioning system.

[0171] Example 7:

[0172] The risk assessment module determines the total risk value, including:

[0173] Preset weighting coefficients are assigned to dead reckoning and predicted future cumulative risk;

[0174] The total risk value is determined by weighted summation of dead reckoning and predicted future cumulative risk.

[0175] This embodiment specifically illustrates how the risk assessment module integrates current and future risks to arrive at the basis for the final decision;

[0176] The risk assessment module determines the total risk value. Its specific implementation includes:

[0177] For deadlock and the predicted future cumulative risks Assign preset weight coefficients ;

[0178] By analyzing deadlock entropy and the predicted future cumulative risks Perform a weighted summation to determine the total risk value. In this embodiment, the calculation formula is:

[0179] ;

[0180] The purpose of this model is to provide a unified risk metric that integrates the present and the future, serving as the sole basis for triggering tiered corrective strategies; when formulating corrective strategies, the degree of current disorientation must also be considered. and the risk of getting lost in the future ;

[0181] in: The total risk value is a dimensionless scalar, which serves as the basis for triggering tiered adjustments in decision-making.

[0182] The current dead reckoning is derived from the dead reckoning calculation steps;

[0183] The accumulated risk for the predicted future originates from the entropy increase prediction step;

[0184] The preset dead reckoning entropy weighting coefficient, These are preset predicted risk weight coefficients, both of which are dimensionless weights derived from pre-set risk preference strategies based on the system, and satisfy the following conditions: ;

[0185] Weight The values ​​are determined based on the system's risk appetite; under the resilience-first strategy of this invention, the system is more sensitive to future risks, therefore... Sensitivity to future risks should be assigned a high value;

[0186] This embodiment addresses the current risks. and future risks A weighted summation was performed to generate a uniform total risk value. This design ensures that subsequent corrective decisions no longer rely solely on... It is not a delayed decision-making process, but rather takes into account... Forward-looking decision-making; and, the scheme allows for adjustments to weights. It allows for flexible configuration of the system's risk preferences, offering greater flexibility and configurability.

[0187] Example 8:

[0188] First-level corrections include:

[0189] The total risk value is determined to be greater than the preset dead reckoning high-risk threshold but less than or equal to the preset dead reckoning failure threshold;

[0190] Actively reduce the robot's forward speed and keep the winding mechanism in a forward state in an attempt to enable the system to self-correct.

[0191] This embodiment specifically illustrates the first-level correction strategy under high-risk conditions;

[0192] The first-level correction is specifically implemented as follows:

[0193] Determine the total risk value The dead reckoning is higher than the preset high-risk threshold. And less than or equal to the preset dead reckoning failure threshold This condition The high-risk state of the system has been defined;

[0194] Triggering corrective action: Actively reduce the robot's forward speed. ;

[0195] Maintain task continuity: and maintain the instructions of the reel mechanism. Forward state In an attempt to enable the system to self-correct;

[0196] When the system is in a high-risk state, the location reliability decreases, but it is not completely lost; the motivation for the first-level correction is to sacrifice some efficiency indicators, such as speed reduction, in exchange for the opportunity to recover the location.

[0197] It actively reduces the robot's forward movement speed. This will lead to actual speed Lower; lower This slows down the accumulation of drift; at the same time, the lower speed provides a more stable signal acquisition window for the hydroacoustic sensor, helping to extract more accurate environmental features. Both of these contribute to reducing dead reckoning. Thus The value fell back to The following steps will enable self-correction;

[0198] The first-level correction defined in this embodiment is a low-cost online correction strategy. It avoids immediately performing costly stop-and-go operations in high-risk situations, and instead intelligently creates recovery conditions for the positioning system by actively slowing down, thereby achieving a balance between ensuring mission continuity and improving positioning resilience.

[0199] Example 9:

[0200] Secondary corrections include:

[0201] The total risk value is determined to be greater than the preset dead reckoning failure threshold;

[0202] Immediately set the forward speed to zero and send a retraction command to the tube winding mechanism;

[0203] When the robot passively retreats to the vicinity of the trusted anchor point, it stops retraction, forcibly resets the positioning system state to the trusted anchor point, and resets the dead reckoning to zero.

[0204] This embodiment specifically illustrates the two-level correction strategy under the failure state, which is the core embodiment of the toughness priority of this invention;

[0205] The second-level correction is specifically implemented as follows:

[0206] Determine the total risk value The failure threshold is greater than the preset dead reckoning threshold. This condition It defines whether the system has entered a failed or lost state;

[0207] Triggering a resilience-first strategy, sacrificing efficiency and coverage, will execute in the following order:

[0208] Immediately set the forward command speed to zero. This stops the further accumulation of error;

[0209] And send a retraction command to the tube winding mechanism. This causes the robot to retreat passively.

[0210] The robot stops retracting when it passively retreats to the vicinity of a trusted anchor point;

[0211] Trusted Anchors This refers to the closed-loop correction module being in a safe state when the system is in a safe state. The last high-confidence position recorded in real time; the determination of backtracking to a nearby location can be achieved by estimating the inertial parameters by integration or by estimating the backtracking time;

[0212] Forcefully reset the positioning system state to a trusted anchor point, i.e., set the current estimated location. And reset the dead reckoning to zero. This allows the positioning system to be reborn from a known safe state;

[0213] when When the positioning system is determined to be lost, the first-level self-correction has failed; at this point, a highly reliable hard reset operation must be performed to ensure the survival of the positioning system.

[0214] Trusted Anchors This is the key to this strategy; because Is The location is recorded in time, and its reliability is extremely high; the essence of the second-level correction is to abandon the current unreliable state and bring the robot back through physical rollback. The corresponding real physical location will determine the internal state of the positioning system. Forcefully restore the system to the safe state corresponding to the anchor point, thereby completing the system's resurrection;

[0215] The secondary correction defined in this embodiment addresses a key industry pain point: namely, the failure of inertial navigation systems after they become lost. The problem of not being able to self-recover; addressed by introducing trusted anchors. With the back-reset mechanism, this invention provides a highly reliable hard reset capability for the positioning system, ensuring that the robot can safely return and reset the task when it encounters an unrecoverable positioning failure, greatly improving the robustness of the system and the safety of operation in extreme environments.

[0216] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A dust removal air duct unblocking robot automatic positioning system, characterized in that, The method comprises the following steps: a data acquisition module is configured to obtain inertial parameters of the robot, hydraulic environment parameters, a preset digital pipeline map, a preset high-risk threshold of dead reckoning entropy, and a preset failure threshold of dead reckoning entropy, wherein the inertial parameters include a three-axis acceleration vector and a three-axis angular velocity vector, and the hydraulic environment parameters include a water flow pressure and a water flow impact acoustic signal; a state estimation module is configured to determine a current estimated position and an estimated velocity based on the three-axis acceleration vector and the three-axis angular velocity vector through a preset dead reckoning model; the state estimation module is further configured to determine a hydraulic feedback feature vector based on the water flow pressure and the water flow impact acoustic signal through a preset feature extraction network; a risk assessment module is configured to calculate a dead reckoning entropy by comprehensively estimating the estimated position, the hydraulic feedback feature vector, and the digital pipeline map; the calculation formula of the dead reckoning entropy is: ； wherein, is a dead reckoning entropy, is an estimated position, is a hydraulic feedback feature vector, is a digital map of the pipe, is a conditional probability of observing the hydraulic feedback feature vector; the risk assessment module is further configured to calculate a physical disturbance factor based on the water flow pressure; the risk assessment module is further configured to determine a predicted future cumulative risk by the preset entropy increase prediction model according to the physical disturbance factor, the dead reckoning entropy, and the estimated velocity; the risk assessment module is further configured to determine a total risk value by combining the dead reckoning entropy and the predicted future cumulative risk; a closed-loop correction module is configured to generate a hierarchical correction instruction based on the total risk value, the dead reckoning entropy, the preset high-risk threshold of dead reckoning entropy, and the preset failure threshold of dead reckoning entropy; the closed-loop correction module is further configured to update the estimated position to a trusted anchor point when the dead reckoning entropy is less than or equal to the preset high-risk threshold of dead reckoning entropy. When the total risk value is greater than the preset high-risk threshold of dead reckoning entropy and less than or equal to the preset failure threshold of dead reckoning entropy, a first-level correction is performed to reduce the speed of the robot forward instruction. When the total risk value is greater than the preset failure threshold of dead reckoning entropy, a second-level correction is performed to set the forward instruction speed to zero, activate the pipe winding mechanism to retract the instruction, and reset the estimated position to the trusted anchor point when the robot retreats to the vicinity of the trusted anchor point.

2. The automatic positioning system for a dust-removal air duct unblocking robot according to claim 1, characterized in that, The state estimation module determines the current estimated position and the estimated velocity based on the three-axis acceleration vector and the three-axis angular velocity vector, comprising: constructing a basic kinematics model of dead reckoning based on the three-axis acceleration vector and the three-axis angular velocity vector; using the state of the last moment to recursively estimate the estimated position and the estimated velocity at the current moment through the basic kinematics model.

3. The automatic positioning system for a dust pipe unblocking robot according to claim 1, wherein The state estimation module determines the hydraulic feedback feature vector based on the water flow pressure and the water flow impact acoustic signal, comprising: inputting the water flow pressure and the water flow impact acoustic signal as time series signals; processing the time series signals through the preset feature extraction network to convert them into a structured hydraulic feedback feature vector.

4. The automatic positioning system for a dust pipe unblocking robot according to claim 1, wherein The risk assessment module calculates the dead reckoning entropy, comprising: calling the estimated position, the hydraulic feedback feature vector, and the digital pipeline map; outputting the conditional probability of observing the hydraulic feedback feature vector at the estimated position on the digital pipeline map through a probability model; determining the dead reckoning entropy by calculating the negative log-likelihood value of the conditional probability.

5. The automatic positioning system for a dust pipe unblocking robot according to claim 1, wherein, The risk assessment module calculates the physical disturbance factor, comprising: real-time monitoring the time series of the water flow pressure; calculating the instantaneous change rate of the water flow pressure; quantitatively determining the physical disturbance factor based on the instantaneous change rate of the water flow pressure and the recent average pressure, and combining a preset sensitivity weight coefficient.

6. The automatic positioning system for a dust-removal air duct unblocking robot according to claim 1, characterized in that, The risk assessment module determines a predicted future cumulative risk, comprising: calling a physical interference factor, a dead reckoning entropy, and an estimated speed; inputting the physical interference factor, the navigation entropy and the estimated speed into a preset entropy increase prediction model, and determining the predicted future cumulative risk through a formula determining a predicted future cumulative risk, wherein, for the predicted future cumulative risk, the preset entropy increase prediction model, the physical interference factor, the estimated speed.

7. The automatic positioning system for a dust pipe unblocking robot according to claim 1, wherein, The risk assessment module determines a total risk value, comprising: assigning a preset weight coefficient to the dead reckoning entropy and the predicted future cumulative risk; The total risk value is determined by a weighted sum of the sailing entropy and the predicted future accumulated risk by the formula wherein, is the total risk value, is a preset sailing entropy weight coefficient, is a preset predicted risk weight coefficient.

8. The automatic positioning system for a dust-removal air duct unblocking robot according to claim 1, characterized in that, The first-level correction, comprising: determining that the total risk value is greater than a preset dead reckoning entropy high-risk threshold and less than or equal to a preset dead reckoning entropy failure threshold; actively reducing the forward instruction speed of the robot and keeping the pipe winding mechanism instruction as a forward state to attempt to make the system self-correct.

9. The automatic positioning system for a dust-removal air duct unblocking robot according to claim 1, characterized in that, The second-level correction, comprising: determining that the total risk value is greater than a preset dead reckoning entropy failure threshold; immediately setting the forward instruction speed to zero and sending a retraction instruction to the pipe winding mechanism; when the robot is passively retracted to the vicinity of a trusted anchor point, stopping retraction and forcibly resetting the positioning system state to the trusted anchor point and resetting the dead reckoning entropy to zero.

Images