An ultrasonic sensor self-adaptive calibration method based on a knowledge graph

Abstract

The application discloses an ultrasonic sensor self-adaptive calibration method based on a knowledge graph, which comprises the following steps: acquiring calibration input data, constructing a calibration knowledge graph and converting the calibration knowledge graph into a graph input structure; constructing an NBFNet inference model based on the graph input structure, including a path relaxation unit, a cost propagation unit and a physical feasible domain constraint; performing physical feasible domain constraint judgment processing and multi-hop cost propagation processing to generate a path propagation state set; constructing a hierarchical propagation structure in the NBFNet inference model; inputting the path propagation state set into the hierarchical propagation structure to perform hierarchical path propagation processing and generate a hierarchical propagation state set; performing forward path inference processing and parameter correction processing to generate a converged calibration parameter set; and writing the converged calibration parameter set into a calibration parameter configuration of an ultrasonic sensor to generate a self-adaptive calibration result. The application improves calibration stability and environmental adaptability.

Description

Technical Field

[0001] This invention relates to the field of sensor calibration and metrology technology, and in particular to an adaptive calibration method for ultrasonic sensors based on knowledge graphs. Background Technology

[0002] Ultrasonic sensors are widely used in distance measurement, liquid level detection, obstacle sensing, and industrial automation due to their simple structure, low cost, and wide range of applicable environments. To ensure measurement accuracy, ultrasonic sensors typically require parameter calibration during use to compensate for measurement errors caused by factors such as changes in sound velocity, system delay, and differences in components.

[0003] In existing technologies, the calibration of ultrasonic sensors mostly relies on manual calibration or parameter correction methods based on a single model, and calibration is usually completed in a fixed environment or during the initial installation phase. These methods struggle to maintain stable accuracy under continuously changing environmental temperatures, humidity, air pressure, and installation conditions, and have limited ability to comprehensively utilize historical calibration data and real-time observation data, easily leading to cumulative biases.

[0004] On the other hand, some improvement schemes attempt to introduce data-driven models to adjust sensor parameters online. However, existing methods mostly focus on numerical fitting or local error correction, lacking systematic modeling of the inherent relationship between sensor structure, environmental influencing factors and observation process. This makes it difficult to accurately depict the error propagation path under the coupling effect of multiple factors, resulting in a lack of interpretability and stability in the calibration process.

[0005] Furthermore, in complex application scenarios, existing calibration techniques often fail to effectively distinguish the contribution of different propagation paths and different influencing factors to measurement deviations, lack a constraint mechanism for the rationality of calibration parameter updates, and are prone to problems such as inconsistent parameter updates or insufficient convergence.

[0006] Therefore, how to provide an adaptive calibration method for ultrasonic sensors based on knowledge graphs is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose an adaptive calibration method for ultrasonic sensors based on a knowledge graph. This invention constructs a calibration knowledge graph that integrates sensor structural information, environmental state information, echo observation information, and historical calibration information. It introduces a path-based graph reasoning model to perform structured modeling and reasoning analysis of the calibration parameter update process, enabling traceable analysis and constraint correction of the sources of calibration deviations. This invention effectively characterizes the error propagation relationship under multi-factor coupling conditions through a hierarchical propagation and path consistency constraint mechanism, and obtains a convergent set of calibration parameters through iterative correction, thereby achieving adaptive calibration of ultrasonic sensors in complex environments. This method has the advantages of high calibration process stability, strong parameter update consistency, and strong adaptability to environmental changes.

[0008] An adaptive calibration method for an ultrasonic sensor based on a knowledge graph, according to an embodiment of the present invention, includes the following steps:

[0009] Acquire calibration input data, construct a calibration knowledge graph, and convert it into a graph input structure;

[0010] The NBFNet inference model is constructed based on the graph input structure, including path relaxation units and cost propagation units, and physical feasible region constraints are also constructed.

[0011] Physical feasible region constraint determination is performed in the path relaxation unit, and multi-hop cost propagation is performed in the cost propagation unit to generate a path propagation state set.

[0012] A hierarchical propagation structure is constructed in the NBFNet inference model, including a calibration-driven propagation layer, an environment modulation propagation layer, and an observation mapping propagation layer;

[0013] The path propagation state set is input into the hierarchical propagation structure to perform hierarchical path propagation processing. Propagation update processing is performed in each of the three layers to generate a hierarchical propagation state set.

[0014] In the NBFNet inference model, forward path inference and parameter correction are performed based on the hierarchical propagation state set to generate a convergence calibration parameter set.

[0015] The convergent calibration parameter set is written into the calibration parameter configuration of the ultrasonic sensor to generate adaptive calibration results.

[0016] Optionally, the calibration input data includes ultrasonic sensor structure data, environmental state data, echo observation data, and historical calibration data. The calibration knowledge graph includes a set of entity nodes and a set of relational edges. The set of entity nodes includes sensor entity nodes, environmental entity nodes, calibration parameter entity nodes, and echo observation entity nodes. The set of relational edges includes calibration-driven relational edges, environmental modulation relational edges, and observation mapping relational edges.

[0017] Optionally, the construction of the NBFNet inference model includes:

[0018] The NBFNet inference model is initialized in the computational framework based on the graph input structure, and path relaxation units and cost propagation units are configured.

[0019] Sound speed range constraints are constructed based on environmental condition data to limit the range of sound speed values.

[0020] Based on echo observation data, a propagation delay constraint is constructed to limit the echo propagation time interval;

[0021] By simultaneously incorporating the sound speed range constraint and the propagation delay constraint into the joint calculation process of the path relaxation unit and the cost propagation unit, a physical feasible region constraint is formed.

[0022] Optionally, the generation of the path propagation state set includes:

[0023] Based on the graph input structure, candidate relation paths consisting of a set of entity nodes and a set of relation edges are enumerated in the path relaxation unit, and the corresponding propagation condition parameters are associated.

[0024] For each candidate relationship path, the propagation condition parameters are judged based on the physical feasible domain constraint. When the propagation condition parameters simultaneously satisfy the sound speed range constraint and the propagation delay constraint, the corresponding candidate relationship path is marked as a physical feasible path and aggregated to form a set of physical feasible paths.

[0025] The set of physically feasible paths is input into the cost propagation unit. Multi-hop cost propagation processing is performed based on the set of physically feasible paths. The cumulative cost of each physically feasible path in the set of physically feasible paths is updated according to the order of the relational paths, and a set of path propagation states is generated.

[0026] Optionally, the construction of the hierarchical propagation structure includes:

[0027] Based on the graph input structure, the relation edge set is structurally partitioned in the NBFNet inference model. According to the relation types recorded in the relation edge set, the relation edge set is divided into calibration-driven relation edge subset, environment-modulated relation edge subset, and observation-mapping relation edge subset.

[0028] A calibration-driven propagation layer is constructed in the NBFNet inference model based on a subset of calibration-driven relational edges, and a propagation update structure for receiving the set of propagation state of the path is configured.

[0029] An environmental modulation propagation layer is constructed based on a subset of environmental modulation relation edges, and a propagation update structure is configured to receive the output results of the calibration-driven propagation layer.

[0030] An observation mapping propagation layer is constructed based on a subset of observation mapping relationship edges, and a propagation update structure is configured to receive the output results of the environment modulation propagation layer.

[0031] A hierarchical propagation structure is formed based on the calibration-driven propagation layer, the environment modulation propagation layer, and the observation mapping propagation layer.

[0032] Optionally, the generation of the hierarchical propagation state set includes:

[0033] The path propagation state set is input into the hierarchical propagation structure, and the propagation hierarchy mapping is performed on the physically feasible path according to the relation type recorded in the relation edge set.

[0034] In the calibration-driven propagation layer, calibration-driven propagation update processing is performed on the path propagation state mapped to the calibration-driven propagation layer based on the calibration-driven relationship edge pairs, generating a set of drive propagation states;

[0035] The driving propagation state set is input into the environment modulation propagation layer. Based on the environment modulation relation edges, the path propagation state in the driving propagation state set is updated by the environment modulation propagation state set to generate the modulation propagation state set.

[0036] The modulation propagation state set is input into the observation mapping propagation layer. Based on the observation mapping relationship edges, the path propagation state in the modulation propagation state set is updated by observation mapping propagation, generating a hierarchical propagation state set.

[0037] Optionally, the generation of the convergence calibration parameter set includes:

[0038] The hierarchical propagation state set is input into the NBFNet inference model, and forward path inference processing is performed on the path states corresponding to the hierarchical propagation state set. Based on the propagation results of each path state in the hierarchical propagation state set, a calibration parameter update set is generated.

[0039] The calibration parameter update set is jointly processed with echo observation data and historical calibration data. Based on the joint processing, calibration deviation calculation is performed on the calibration parameter update set to generate calibration deviation results.

[0040] Based on the calibration deviation results, reverse path backtracking is performed on the path states corresponding to the calibration parameter update set in the graph input structure to generate a backtracking path set.

[0041] Based on the backtracking path set, path consistency constraint processing is performed on the calibration parameter update set, and constraint correction is performed on the calibration parameter update results that do not meet the path consistency constraint requirements;

[0042] The output of the path consistency constraint processing is used to perform iterative correction processing on the calibration parameter update set until the calibration parameter update set meets the preset convergence condition, thus generating a converged calibration parameter set.

[0043] Optionally, the generation of the adaptive calibration result includes:

[0044] The convergence calibration parameter set is mapped to the calibration parameter configuration of the ultrasonic sensor. Based on the predefined parameter identifier relationship in the calibration parameter configuration, the configuration writing position corresponding to each calibration parameter in the convergence calibration parameter set is determined.

[0045] Based on the result of parameter item mapping processing, the calibration parameters in the converged calibration parameter set are written to the calibration parameter configuration of the ultrasonic sensor according to the corresponding configuration writing position, so that the calibration parameter configuration is updated to a parameter state consistent with the converged calibration parameter set.

[0046] After completing the writing of the calibration parameter configuration, the updated calibration parameter configuration is used as the current calibration parameter state of the ultrasonic sensor, and adaptive calibration results are generated.

[0047] The beneficial effects of this invention are:

[0048] First, this invention introduces a knowledge graph to uniformly model the relationships between ultrasonic sensor structural information, environmental state information, echo observation information, and historical calibration information. This transforms the calibration process from traditional parameter-level correction to a holistic reasoning process based on relationships and paths, thereby systematically characterizing the error propagation mechanism under the coupling effect of multiple factors and improving the stability and consistency of calibration results.

[0049] Secondly, this invention uses a graph reasoning model to perform path-level reasoning and hierarchical propagation processing on the calibration parameter update process, so that the influence of different types of relationships on the calibration parameters can be distinguished in an orderly manner and transmitted layer by layer. This effectively avoids the problem of unreasonable parameter updates caused by environmental changes or observation noise in the prior art, and enhances the controllability and convergence of the calibration process.

[0050] Furthermore, by introducing a path backtracking and path consistency constraint mechanism, this invention performs reverse analysis on the sources of calibration deviation and constrains and corrects the calibration parameter update results, enabling the calibration parameters to converge stably after multiple iterations. This achieves adaptive calibration of ultrasonic sensors under complex working conditions and long-term operating conditions, significantly improving the sensor's measurement accuracy and operational reliability. Attached Figure Description

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0052] Figure 1 This is an overall flowchart of an adaptive calibration method for ultrasonic sensors based on knowledge graphs proposed in this invention.

[0053] Figure 2 This is a schematic diagram of the structure of the present invention, which constructs an NBFNet inference model based on a calibration knowledge graph and performs physical feasible region constraints and path propagation processing.

[0054] Figure 3 This is a schematic diagram illustrating the execution path backtracking and path consistency constraints based on the hierarchical propagation structure in this invention. Detailed Implementation

[0055] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0056] refer to Figure 1-3 An adaptive calibration method for ultrasonic sensors based on knowledge graphs includes the following steps:

[0057] Acquire calibration input data, which includes ultrasonic sensor structure data, environmental state data, echo observation data, and historical calibration data. Construct a calibration knowledge graph based on the calibration input data. The calibration knowledge graph includes a set of entity nodes and a set of relation edges. The set of entity nodes includes sensor entity nodes, environmental entity nodes, calibration parameter entity nodes, and echo observation entity nodes. The set of relation edges includes calibration-driven relation edges, environmental modulation relation edges, and observation mapping relation edges. Convert the calibration knowledge graph into a graph input structure.

[0058] An NBFNet inference model is constructed based on a graph input structure. The NBFNet inference model includes path relaxation units and cost propagation units. Physical feasible region constraints are constructed in the NBFNet inference model based on environmental state data and echo observation data. The physical feasible region constraints include sound speed range constraints and propagation delay constraints.

[0059] In the path relaxation unit, the physical feasible region constraint determination process is performed on the relational paths of the graph input structure to obtain a set of physical feasible paths. In the cost propagation unit, multi-hop cost propagation processing is performed based on the set of physical feasible paths to generate a set of path propagation states.

[0060] A hierarchical propagation structure is constructed in the NBFNet inference model, which includes a calibration-driven propagation layer, an environment modulation propagation layer, and an observation-mapping propagation layer.

[0061] Based on the relation types recorded in the relation edge set, the path propagation state set is input into the hierarchical propagation structure to perform hierarchical path propagation processing. In the calibration-driven propagation layer, calibration-driven propagation update processing is performed on the path propagation state set based on calibration-driven relation edges to generate a drive propagation state set. In the environmental modulation propagation layer, environmental modulation propagation update processing is performed on the drive propagation state set based on environmental modulation relation edges to generate a modulation propagation state set. In the observation mapping propagation layer, observation mapping propagation update processing is performed on the modulation propagation state set based on observation mapping relation edges to generate a hierarchical propagation state set.

[0062] In the NBFNet inference model, forward path inference is performed based on the hierarchical propagation state set to obtain the calibration parameter update set; calibration deviation calculation is performed based on the calibration parameter update set, echo observation data, and historical calibration data to obtain the calibration deviation result; reverse path backtracking is performed on the graph input structure based on the calibration deviation result to obtain the backtracking path set; path consistency constraint processing is performed on the calibration parameter update set based on the backtracking path set; and iterative correction processing is performed on the calibration parameter update set based on the output of the path consistency constraint processing to generate a converged calibration parameter set.

[0063] The convergent calibration parameter set is written into the calibration parameter configuration of the ultrasonic sensor to generate adaptive calibration results.

[0064] In this embodiment, the calibration input data includes ultrasonic sensor structure data, environmental state data, echo observation data, and historical calibration data. The calibration knowledge graph includes a set of entity nodes and a set of relational edges. The set of entity nodes includes sensor entity nodes, environmental entity nodes, calibration parameter entity nodes, and echo observation entity nodes. The set of relational edges includes calibration-driven relational edges, environmental modulation relational edges, and observation mapping relational edges. The process of converting the calibration knowledge graph into a graph input structure specifically involves mapping the entity nodes in the calibration knowledge graph to nodes in the graph structure, mapping the relational edges to directed edges in the graph structure, and configuring feature vectors corresponding to the calibration input data for the nodes and edges, respectively, to form a graph input structure that can be used by the NBFNet inference model.

[0065] In this embodiment, the construction of the NBFNet inference model includes:

[0066] The NBFNet inference model is initialized in the computation framework based on the graph input structure, and path relaxation unit and cost propagation unit are configured. The path relaxation unit is used to perform layer-by-layer relaxation calculation on the relational paths in the graph input structure, and the cost propagation unit is used to perform propagation update on the cumulative cost of the path during the path relaxation calculation.

[0067] Based on environmental state data, a sound speed range constraint is constructed in the NBFNet inference model. The sound speed range constraint is applied to the path relaxation unit by limiting the range of sound speed values, and is used to constrain the propagation conditions related to sound speed in the path relaxation calculation.

[0068] The process of constructing the sound speed range constraint includes: inputting environmental state data into the NBFNet inference model; determining the theoretical range of sound speed corresponding to the current environmental state based on the environmental temperature, humidity, and air pressure information reflected in the environmental state data; and representing this theoretical range of sound speed as the sound speed range constraint in the NBFNet inference model. When performing relational path relaxation calculation in the path relaxation unit, propagation condition parameters related to sound speed are associated with each candidate relational path. During the path relaxation calculation, it is determined whether the propagation condition parameters are within the theoretical range of sound speed defined by the sound speed range constraint. When the propagation condition parameters are within the theoretical range of sound speed, the corresponding relational path is allowed to participate in the path relaxation calculation; when the propagation condition parameters exceed the theoretical range of sound speed, the corresponding relational path is prohibited from participating in the path relaxation calculation. Thus, the propagation of relational paths in the path relaxation unit is restricted through the sound speed range constraint.

[0069] Based on echo observation data, a propagation delay constraint is constructed in the NBFNet inference model. The propagation delay constraint is applied to the cost propagation unit by limiting the echo propagation time interval, which is used to constrain the propagation update process of the cumulative cost of the path.

[0070] The process of constructing propagation delay constraints includes: inputting echo observation data into the NBFNet inference model; calculating the corresponding echo propagation time based on the ultrasonic wave transmission time and echo reception time information recorded in the echo observation data; determining the effective value range of the echo propagation time based on multiple echo observation data; representing this effective value range of the echo propagation time as a propagation delay constraint in the NBFNet inference model; when the cost propagation unit performs path cumulative cost propagation update, associating a propagation time cost value with each relational path; and determining whether the propagation time cost value is within the effective value range of the echo propagation time defined by the propagation delay constraint during the cost propagation update process; when the propagation time cost value is within the effective value range of the echo propagation time, the corresponding relational path is allowed to perform cost propagation update; when the propagation time cost value exceeds the effective value range of the echo propagation time, the corresponding relational path is prohibited from performing cost propagation update, thereby limiting the path cumulative cost propagation in the cost propagation unit through propagation delay constraints.

[0071] By simultaneously introducing sound speed range constraints and propagation delay constraints into the joint computation process of path relaxation units and cost propagation units, physical feasible region constraints are formed. In the path inference computation of the NBFNet inference model, relational paths in the graph input structure are filtered and updated based on the physical feasible region constraints.

[0072] In this embodiment, the generation of the path propagation state set includes:

[0073] Based on the graph input structure, candidate relation paths consisting of a set of entity nodes and a set of relation edges are enumerated in the path relaxation unit, and corresponding propagation condition parameters are associated with each candidate relation path.

[0074] For each candidate relationship path, the propagation condition parameters are judged based on the physical feasible domain constraint. When the propagation condition parameters simultaneously satisfy the sound speed range constraint and the propagation delay constraint, the corresponding candidate relationship path is marked as a physical feasible path, and all marked physical feasible paths are aggregated to form a physical feasible path set.

[0075] The set of physically feasible paths is input into the cost propagation unit. In the cost propagation unit, multi-hop cost propagation processing is performed based on the set of physically feasible paths. For each physically feasible path in the set of physically feasible paths, the cumulative cost of the path is updated according to the order of the relational paths, and a set of path propagation states is generated.

[0076] In this embodiment, the construction of the hierarchical propagation structure includes:

[0077] Based on the graph input structure, the relation edge set is structurally partitioned in the NBFNet inference model. According to the relation types recorded in the relation edge set, the relation edge set is divided into calibration-driven relation edge subset, environment-modulated relation edge subset, and observation-mapping relation edge subset.

[0078] A calibration-driven propagation layer is constructed in the NBFNet inference model based on a subset of calibration-driven relation edges. A propagation update structure for receiving the set of propagation states of the path is configured in the calibration-driven propagation layer, so that the calibration-driven propagation layer only performs propagation updates on the relation paths corresponding to the subset of calibration-driven relation edges.

[0079] An environment modulation propagation layer is constructed in the NBFNet inference model based on the subset of environment modulation relation edges. A propagation update structure for receiving the output results of the calibration-driven propagation layer is configured in the environment modulation propagation layer, so that the environment modulation propagation layer only performs propagation update on the relation paths corresponding to the subset of environment modulation relation edges.

[0080] An observation mapping propagation layer is constructed in the NBFNet inference model based on the observation mapping relation edge subset. A propagation update structure for receiving the output results of the environment modulation propagation layer is configured in the observation mapping propagation layer, so that the observation mapping propagation layer only performs propagation update on the relation paths corresponding to the observation mapping relation edge subset.

[0081] It consists of a layered propagation structure comprising a calibration-driven propagation layer, an environmental modulation propagation layer, and an observation mapping propagation layer.

[0082] In this embodiment, the generation of the hierarchical propagation state set includes:

[0083] The path propagation state set is input into the hierarchical propagation structure, and the physical feasible paths in the path propagation state set are mapped according to the relationship types recorded in the relationship edge set, so that each relationship path corresponds to the calibration-driven propagation layer, the environment modulation propagation layer, and the observation mapping propagation layer.

[0084] In the calibration-driven propagation layer, calibration-driven propagation update processing is performed on the path propagation state mapped to the calibration-driven propagation layer based on the calibration-driven relationship edge pairs, generating a set of drive propagation states;

[0085] The calibration-driven propagation update process is used to introduce structural influence information related to calibration parameters into the calibration-driven propagation layer. Specifically, within the calibration-driven propagation layer, based on the parameter dependency structure represented by the calibration-driven relationship edge, a targeted update is performed on the path propagation state mapped to the propagation layer, so that the path propagation state explicitly reflects the driving correlation between calibration parameters, and a set of driving propagation states for subsequent environmental modulation propagation is formed through aggregation processing.

[0086] The driving propagation state set is input into the environment modulation propagation layer. In the environment modulation propagation layer, the path propagation state in the driving propagation state set is updated based on the environment modulation relation edge to generate the modulation propagation state set.

[0087] The environmental modulation propagation update process is used to correct the environmental factors of the driving propagation state set in the environmental modulation propagation layer. Specifically, in the environmental modulation propagation layer, based on the environmental influence mechanism described by the environmental modulation relationship edge, the path propagation state that already contains calibration driving information is modulated and updated so that the path propagation state further integrates the influence of the environmental state on the propagation characteristics, and the modulation propagation state set for observation mapping propagation is generated through aggregation processing.

[0088] The modulation propagation state set is input into the observation mapping propagation layer. In the observation mapping propagation layer, the path propagation state in the modulation propagation state set is updated based on the observation mapping relationship edge to generate a hierarchical propagation state set.

[0089] The observation mapping propagation update process is used to complete the mapping alignment between path propagation state and observation information in the observation mapping propagation layer. Specifically, in the observation mapping propagation layer, based on the observation association structure depicted by the observation mapping relationship edge, the path propagation state in the modulation propagation state set is mapped and updated so that the path propagation state is consistent with the echo observation characteristics in a unified representation space, thereby forming a hierarchical propagation state set for subsequent path inference.

[0090] In this embodiment, the generation of the convergence calibration parameter set includes:

[0091] The hierarchical propagation state set is input into the NBFNet inference model. In the NBFNet inference model, forward path inference processing is performed on the path states corresponding to the hierarchical propagation state set. Based on the propagation results of each path state in the hierarchical propagation state set, a calibration parameter update set is generated.

[0092] The forward path reasoning process is a process of forward reasoning mapping of calibration parameters based on the path propagation results recorded in the hierarchical propagation state set. Specifically, after receiving the hierarchical propagation state set, the NBFNet inference model performs convergence processing on the path propagation states associated with the calibration parameter entity nodes based on the comprehensive propagation results of the calibration driving relationship, environmental modulation relationship and observation mapping relationship represented by each path propagation state in the hierarchical propagation state set. The converged path propagation information is then mapped to the corresponding calibration parameter space, thereby forming a calibration parameter update set that reflects the current path propagation results. This set is used to characterize the update direction and update magnitude of the calibration parameters under the current propagation state conditions.

[0093] The calibration parameter update set is jointly processed with echo observation data and historical calibration data. Based on the joint processing, calibration deviation calculation is performed on the calibration parameter update set to generate calibration deviation results.

[0094] The calibration deviation calculation process is a process of evaluating the deviation of the calibration parameter update set based on the data correspondence established by joint processing, without writing the calibration parameter configuration of the ultrasonic sensor. Specifically, it includes: extracting the echo propagation time information corresponding to the current propagation path from the echo observation data, and extracting the historical echo propagation time reference information corresponding to the same propagation path from the historical calibration data; aligning and matching the echo propagation time information with the historical echo propagation time reference information to form a comparison benchmark for deviation evaluation; performing deviation evaluation calculation on the comparison benchmark according to the update direction and update amplitude recorded in the calibration parameter update set to obtain the deviation amount representing the current update direction and update amplitude in the echo propagation time dimension; and collecting the deviation amounts corresponding to each propagation path to form the calibration deviation result, which is used for subsequent reverse path backtracking processing on the graph input structure.

[0095] Based on the calibration deviation results, reverse path backtracking is performed on the path states corresponding to the calibration parameter update set in the graph input structure to generate a backtracking path set, which is used to characterize the path propagation correlation that leads to the calibration deviation.

[0096] The reverse path backtracking process is a process of reversing the path propagation relationship in the graph input structure based on the calibration deviation result. Specifically, it includes: taking the high deviation path propagation state corresponding to the calibration deviation result as the backtracking starting point in the graph input structure, performing hop-by-hop backtracking along the connection direction of the relation edges recorded in the relation edge set, and identifying the upstream path state that has a propagation dependency relationship with the backtracking starting point hop-by-hop; during the hop-by-hop backtracking process, the deviation contribution of each hop backtracking path state is filtered according to the calibration deviation result, so that the path state that only has a contribution relationship with the calibration deviation result enters the subsequent backtracking; after completing the hop-by-hop backtracking, the backtracking links filtered by the deviation contribution are collected to form a backtracking path set, which is used to characterize the path propagation relationship that caused the calibration deviation and to provide the backtracking path basis for subsequent path consistency constraint processing;

[0097] Based on the backtracking path set, path consistency constraint processing is performed on the calibration parameter update set, and constraint correction is performed on the calibration parameter update results that do not meet the path consistency constraint requirements;

[0098] The path consistency constraint is a constraint used during the correction of the calibration parameter update set to ensure that the calibration parameter update results are consistent with the path propagation relationship represented by the backtracking path set. Specifically, when correcting the calibration parameter update set, the calibration parameter update results are only allowed to change along the propagation directions of the paths included in the backtracking path set, and the relationship between the changes in the calibration parameter update results must match the propagation dependency relationship of each path in the backtracking path set. When the direction or magnitude of change in the calibration parameter update results is inconsistent with the path propagation relationship depicted by the backtracking path set, a constraint correction is applied to the corresponding calibration parameter update results to maintain consistency in the propagation relationship at the path level.

[0099] The output of the path consistency constraint processing is used to perform iterative correction processing on the calibration parameter update set until the calibration parameter update set meets the preset convergence condition, and a converged calibration parameter set is generated.

[0100] The iterative correction process specifically involves iteratively updating the output of the path consistency constraint processing, including: using the output of the path consistency constraint processing as the initial calibration parameter update set; performing consistency verification on the initial calibration parameter update set based on the backtracking path set to determine calibration parameter update results that do not meet the path consistency constraint requirements; performing directional correction processing on the calibration parameter update results that do not meet the path consistency constraint requirements to obtain the corrected calibration parameter update set; returning the corrected calibration parameter update set as the input for the next round of path consistency constraint processing, and repeating the consistency verification and directional correction processing until the calibration parameter update set meets the preset convergence condition, generating a converged calibration parameter set.

[0101] In this embodiment, the generation of the adaptive calibration result includes:

[0102] The convergence calibration parameter set is mapped to the calibration parameter configuration of the ultrasonic sensor. Based on the predefined parameter identifier relationship in the calibration parameter configuration, the configuration writing position corresponding to each calibration parameter in the convergence calibration parameter set is determined.

[0103] Based on the result of parameter item mapping processing, the calibration parameters in the converged calibration parameter set are written to the calibration parameter configuration of the ultrasonic sensor according to the corresponding configuration writing position, so that the calibration parameter configuration is updated to a parameter state consistent with the converged calibration parameter set.

[0104] After the calibration parameter configuration is written, the updated calibration parameter configuration is used as the current calibration parameter state of the ultrasonic sensor, and an adaptive calibration result is generated based on the current calibration parameter state.

[0105] The adaptive calibration result generation process, after determining the updated calibration parameter configuration as the current calibration parameter state of the ultrasonic sensor, includes: using the current calibration parameter state as the effective operating parameter benchmark for the ultrasonic sensor, enabling the ultrasonic sensor to perform signal processing according to the current calibration parameter state during subsequent signal transmission and echo reception; confirming the validity of the echo measurement results output by the ultrasonic sensor under the current calibration parameter state, and associating the current calibration parameter state with the corresponding echo observation performance; when the current calibration parameter state can maintain consistency with the predetermined echo observation requirements, outputting the current calibration parameter state along with the corresponding calibration identification information as the adaptive calibration result, which characterizes that the ultrasonic sensor has completed adaptive calibration for the current environmental state and operating conditions.

[0106] Example 1:

[0107] To verify the feasibility of this invention in practice, it was applied to a real-world ultrasonic sensor ranging system. This system includes multiple ultrasonic sensors that operate in environments with continuously changing conditions, including variations in temperature, humidity, and airflow. In such scenarios, the propagation speed and path of ultrasonic waves are easily affected by the environment, leading to systematic deviations in sensor ranging results. Traditional calibration methods relying on fixed parameters or periodic manual calibration are insufficient to maintain long-term stable accuracy, and the calibration process makes limited use of historical and real-time observation data, resulting in a lack of targeted updates to calibration parameters.

[0108] In this application scenario, sensor structural information, environmental state data, echo observation data, and historical calibration data are used as calibration input data, which are continuously collected and updated. Based on this data, a calibration knowledge graph is constructed, expressing the relationships between sensor entities, environmental entities, calibration parameter entities, and echo observation entities in a structured form. During system operation, the calibration knowledge graph evolves synchronously with data updates, reflecting the changing relationships between various factors under the current operating state.

[0109] In practical applications, the calibration knowledge graph is converted into a graph input structure and input into the NBFNet inference model. Sound speed range constraints and propagation delay constraints are introduced within the model to ensure that the inference process remains within physically feasible limits, preventing paths that do not conform to propagation laws from participating in the calculation. Subsequently, physically feasible paths are selected through path relaxation and cost propagation processes, generating a set of path propagation states. Based on this, a hierarchical propagation structure is constructed by combining calibration-driven relationships, environmental modulation relationships, and observation mapping relationships to perform hierarchical propagation processing of path propagation states, enabling the influence of different types of factors on calibration parameters to be modeled separately.

[0110] During system operation, the model periodically outputs an updated set of calibration parameters and calculates the calibration deviation by combining echo observation data with historical calibration data. The source of the deviation is analyzed through reverse path backtracking, and path consistency constraints are introduced to correct the calibration parameter update results, ensuring stable convergence characteristics in the parameter update process. The final converged calibration parameter set is written into the ultrasonic sensor's calibration parameter configuration and used as the currently valid calibration parameter state.

[0111] Under continuous operating conditions, a comparative test was conducted between the system employing the method of this invention and a system employing a traditional calibration method. During the test, both methods were run under identical environmental and sensor hardware conditions. The calibration effectiveness was compared and analyzed by statistically analyzing multiple rounds of ranging results, changes in calibration parameters, and long-term stability indicators. The experimental data are summarized in the table below.

[0112] Table 1. Comparison of Ultrasonic Sensor Calibration Performance under Different Calibration Methods

[0113] Performance indicators Traditional calibration methods Method of the present invention Average ranging error (mm) 18.6 12.9 Standard deviation of ranging error (mm) 7.4 4.8 Parameter drift magnitude (percentage) 6.2 3.9 Calibration convergence rounds (times) 9 6 Long-term stability rate (percentage) 85.3 89.6

[0114] As shown in Table 1, under the same operating conditions, the method of this invention reduces the average ranging error compared to traditional calibration methods, and the error fluctuation amplitude is significantly reduced, indicating that the calibration results are more stable under different environmental conditions. Simultaneously, the drift amplitude of calibration parameters during long-term operation is controlled, demonstrating that the calibration parameter correction mechanism based on path backtracking and consistency constraints can effectively suppress accumulated deviations. The reduction in calibration convergence rounds reflects a more targeted calibration parameter update process, avoiding invalid or repeated parameter adjustments. The improved long-term operational stability indicates that the system can maintain relatively reliable ranging performance under continuously changing environmental conditions. By introducing a knowledge graph and a path-based reasoning calibration mechanism, this invention enables the ultrasonic sensor calibration process to comprehensively utilize multi-source information and perform structured reasoning, achieving a more stable and reasonable adaptive calibration effect under complex environments and long-term operating conditions, verifying the feasibility and practical value of this invention in real-world applications.

[0115] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An adaptive calibration method for ultrasonic sensors based on knowledge graphs, characterized in that, Includes the following steps: Acquire calibration input data, construct a calibration knowledge graph, and convert it into a graph input structure; The NBFNet inference model is constructed based on the graph input structure, including path relaxation units and cost propagation units, and physical feasible region constraints are also constructed. Physical feasible region constraint determination is performed in the path relaxation unit, and multi-hop cost propagation is performed in the cost propagation unit to generate a path propagation state set. A hierarchical propagation structure is constructed in the NBFNet inference model, including a calibration-driven propagation layer, an environment modulation propagation layer, and an observation mapping propagation layer; The path propagation state set is input into the hierarchical propagation structure to perform hierarchical path propagation processing. Propagation update processing is performed in each of the three layers to generate a hierarchical propagation state set. In the NBFNet inference model, forward path inference and parameter correction are performed based on the hierarchical propagation state set to generate a convergence calibration parameter set. The convergent calibration parameter set is written into the calibration parameter configuration of the ultrasonic sensor to generate adaptive calibration results.

2. The adaptive calibration method for ultrasonic sensors based on knowledge graphs according to claim 1, characterized in that, The calibration input data includes ultrasonic sensor structure data, environmental status data, echo observation data, and historical calibration data. The calibration knowledge graph includes a set of entity nodes and a set of relational edges. The set of entity nodes includes sensor entity nodes, environmental entity nodes, calibration parameter entity nodes, and echo observation entity nodes. The set of relational edges includes calibration-driven relational edges, environmental modulation relational edges, and observation mapping relational edges.

3. The adaptive calibration method for ultrasonic sensors based on knowledge graphs according to claim 1, characterized in that, The construction of the NBFNet inference model includes: The NBFNet inference model is initialized in the computational framework based on the graph input structure, and path relaxation units and cost propagation units are configured. Sound speed range constraints are constructed based on environmental condition data to limit the range of sound speed values. Based on echo observation data, a propagation delay constraint is constructed to limit the echo propagation time interval; By simultaneously incorporating the sound speed range constraint and the propagation delay constraint into the joint calculation process of the path relaxation unit and the cost propagation unit, a physical feasible region constraint is formed.

4. The adaptive calibration method for ultrasonic sensors based on knowledge graphs according to claim 1, characterized in that, The generation of the path propagation state set includes: Based on the graph input structure, candidate relation paths consisting of a set of entity nodes and a set of relation edges are enumerated in the path relaxation unit, and the corresponding propagation condition parameters are associated. For each candidate relationship path, the propagation condition parameters are judged based on the physical feasible domain constraint. When the propagation condition parameters simultaneously satisfy the sound speed range constraint and the propagation delay constraint, the corresponding candidate relationship path is marked as a physical feasible path and aggregated to form a set of physical feasible paths. The set of physically feasible paths is input into the cost propagation unit. Multi-hop cost propagation processing is performed based on the set of physically feasible paths. The cumulative cost of each physically feasible path in the set of physically feasible paths is updated according to the order of the relational paths, and a set of path propagation states is generated.

5. The adaptive calibration method for ultrasonic sensors based on knowledge graphs according to claim 1, characterized in that, The construction of the hierarchical propagation structure includes: Based on the graph input structure, the relation edge set is structurally partitioned in the NBFNet inference model. According to the relation types recorded in the relation edge set, the relation edge set is divided into calibration-driven relation edge subset, environment-modulated relation edge subset, and observation-mapping relation edge subset. A calibration-driven propagation layer is constructed in the NBFNet inference model based on a subset of calibration-driven relational edges, and a propagation update structure is configured to receive the set of propagation state of the path. An environmental modulation propagation layer is constructed based on a subset of environmental modulation relation edges, and a propagation update structure is configured to receive the output results of the calibration-driven propagation layer. An observation mapping propagation layer is constructed based on a subset of observation mapping relationship edges, and a propagation update structure is configured to receive the output results of the environment modulation propagation layer. A hierarchical propagation structure is formed based on the calibration-driven propagation layer, the environment modulation propagation layer, and the observation mapping propagation layer.

6. The adaptive calibration method for ultrasonic sensors based on knowledge graphs according to claim 1, characterized in that, The generation of the hierarchical propagation state set includes: The path propagation state set is input into the hierarchical propagation structure, and the propagation hierarchy mapping is performed on the physically feasible path according to the relation type recorded in the relation edge set. In the calibration-driven propagation layer, calibration-driven propagation update processing is performed on the path propagation state mapped to the calibration-driven propagation layer based on the calibration-driven relationship edge pairs, generating a set of drive propagation states; The driving propagation state set is input into the environment modulation propagation layer. Based on the environment modulation relation edges, the path propagation state in the driving propagation state set is updated by the environment modulation propagation state set to generate the modulation propagation state set. The modulation propagation state set is input into the observation mapping propagation layer. Based on the observation mapping relationship edges, the path propagation state in the modulation propagation state set is updated by observation mapping propagation, generating a hierarchical propagation state set.

7. The adaptive calibration method for ultrasonic sensors based on knowledge graphs according to claim 1, characterized in that, The generation of the convergence calibration parameter set includes: The hierarchical propagation state set is input into the NBFNet inference model, and forward path inference processing is performed on the path states corresponding to the hierarchical propagation state set. Based on the propagation results of each path state in the hierarchical propagation state set, a calibration parameter update set is generated. The calibration parameter update set is jointly processed with echo observation data and historical calibration data. Based on the joint processing, calibration deviation calculation is performed on the calibration parameter update set to generate calibration deviation results. Based on the calibration deviation results, reverse path backtracking is performed on the path states corresponding to the calibration parameter update set in the graph input structure to generate a backtracking path set. Based on the backtracking path set, path consistency constraint processing is performed on the calibration parameter update set, and constraint correction is performed on the calibration parameter update results that do not meet the path consistency constraint requirements; The output of the path consistency constraint processing is used to perform iterative correction processing on the calibration parameter update set until the calibration parameter update set meets the preset convergence condition, thus generating a converged calibration parameter set.

8. The adaptive calibration method for ultrasonic sensors based on knowledge graphs according to claim 1, characterized in that, The generation of the adaptive calibration results includes: The convergence calibration parameter set is mapped to the calibration parameter configuration of the ultrasonic sensor. Based on the predefined parameter identifier relationship in the calibration parameter configuration, the configuration writing position corresponding to each calibration parameter in the convergence calibration parameter set is determined. Based on the result of parameter item mapping processing, the calibration parameters in the converged calibration parameter set are written to the calibration parameter configuration of the ultrasonic sensor according to the corresponding configuration writing position, so that the calibration parameter configuration is updated to a parameter state consistent with the converged calibration parameter set. After completing the writing of the calibration parameter configuration, the updated calibration parameter configuration is used as the current calibration parameter state of the ultrasonic sensor, and adaptive calibration results are generated.

Images