Intelligent detection method for elevator landing door gap safety

Abstract

The application relates to the technical field of elevator safety detection, and provides an intelligent detection method for elevator landing door gap safety, which comprises the following steps: acquiring the static apparent features of the bottom of a door leaf, and preliminarily screening a plurality of candidate force applying areas based on the static apparent features to obtain preliminarily screened candidate areas; applying a micro load smaller than a formal detection force to the preliminarily screened candidate areas and acquiring dynamic mechanical responses, so as to determine the most unfavorable force applying point; applying the formal detection force to the most unfavorable force applying point and synchronously acquiring non-contact and contact gap data; and cross-verification and fusion are carried out based on the two types of gap data to obtain a final gap measurement result. The application realizes objective and accurate positioning of the most unfavorable force applying point and high-reliability measurement of the gap, and improves the detection accuracy and robustness.

Description

Technical Field

[0001] This application relates to the field of elevator safety inspection technology, specifically to an intelligent detection method for elevator door gap safety. Background Technology

[0002] In the field of elevator safety inspection, the ability of landing doors to remain closed under specified testing forces is a key indicator for assessing elevator operational safety. Existing landing door gap detection methods typically rely on manual experience to select the force application point, apply the testing force using a spring tension gauge, and then measure the door gap using contact tools such as feeler gauges or vernier calipers. However, this traditional method has significant drawbacks: firstly, the selection of the force application point is highly dependent on the inspector's subjective experience, making it difficult to objectively and accurately identify the most unfavorable stress point on the landing door structure. This is especially true in complex conditions such as deformation, corrosion, or obstruction at the bottom of the door leaf, easily leading to missed weak areas and poor consistency and reproducibility of the test results. Secondly, the single contact-based gap measurement method is susceptible to human error, oil contamination, and contact pressure, and it is difficult to capture dynamic changes in the gap in real time during force application, resulting in insufficient robustness and accuracy of the measurement data. Therefore, there is an urgent need for an intelligent detection method for elevator landing door gap safety that can objectively locate the most unfavorable stress point and achieve highly reliable gap measurement. Summary of the Invention

[0003] To address the problems of subjective positioning of the most unfavorable force application point and poor anti-interference ability of gap measurement in existing technologies, this application proposes an intelligent detection method for elevator door gap safety, which achieves objective and accurate positioning of the most unfavorable force application point and highly reliable gap measurement.

[0004] To achieve the above objectives, this application adopts the following technical solution: An intelligent detection method for elevator landing door gap safety includes: acquiring static appearance features of the bottom of the door leaf; performing preliminary screening of multiple candidate force application areas on the bottom of the door leaf based on the static appearance features to obtain preliminary candidate areas; applying a micro-load to the preliminary candidate areas and acquiring the dynamic mechanical response of the preliminary candidate areas under the micro-load; determining the most unfavorable force application point from the preliminary candidate areas based on the dynamic mechanical response; wherein the force value of the micro-load is less than the formal detection force; applying the formal detection force to the most unfavorable force application point, and simultaneously acquiring non-contact gap data and contact gap data during the force application process; and performing cross-validation and fusion based on the non-contact gap data and the contact gap data to obtain the final gap measurement result.

[0005] As one implementation method, the preliminary screening of multiple candidate force application areas at the bottom of the door leaf based on the static appearance features to obtain preliminary candidate areas includes: extracting weak feature parameters characterizing the weakness of the bottom structure of the door leaf; standardizing the weak feature parameters and inputting the standardized weak feature parameters into a multi-feature fusion scoring model to obtain the most unfavorable score for each candidate force application area; and selecting a preset number of candidate force application areas as preliminary candidate areas according to the most unfavorable scores from high to low.

[0006] This implementation transforms the approach from qualitative observation to quantitative assessment by quantifying multi-dimensional structural weakness characteristics into a unified worst-case score. This ensures that the initial screening candidate area can fully cover potential structural weakness regions, providing a reliable candidate set for subsequent mechanical fine screening.

[0007] As one implementation, the weak feature parameters include: local pixel width of the door gap, continuity of the bottom edge contour of the door leaf, suspension geometry of the door leaf relative to the lower threshold baseline, area of ​​the accessible region, reachability of the force application end, deformation curvature of the bottom edge of the door leaf, probability of foreign objects in the gap area, quantified value of the degree of corrosion of the bottom edge of the door leaf, rate of change of curvature of the bottom edge contour of the door leaf, and confidence in the detection of foreign objects inside the door gap; the multi-feature fusion scoring model is a gradient boosting regression tree model or a fully connected neural network model.

[0008] This implementation further refines the specific dimensions for characterizing structural weakness, covering multiple aspects such as geometric deformation, surface condition, foreign object interference, and the feasibility of applying force. Combined with the nonlinear fitting capability of the machine learning model, it improves the sensitivity and accuracy of identifying the most unfavorable force application area under complex working conditions.

[0009] As one implementation method, applying a micro-load to the preliminary screening candidate area and obtaining the dynamic mechanical response of the preliminary screening candidate area under the micro-load, and determining the most unfavorable force application point from the preliminary screening candidate area based on the dynamic mechanical response, includes: measuring the micro-displacement change of the door gap in real time during the process of applying the micro-load to the preliminary screening candidate area; using the ratio of the door gap increment to the corresponding force increment as the gap change sensitivity; and selecting the preliminary screening candidate area with the highest gap change sensitivity as the most unfavorable force application point.

[0010] This implementation method utilizes micro-loads to excite the elastic deformation response of the door leaf, and objectively characterizes the mechanical weakness of the structure by calculating the sensitivity to gap changes. This avoids the damage risk that may be caused by directly applying formal testing force, while ensuring that the final selected force application point is indeed the most unfavorable point in a mechanical sense.

[0011] As one implementation, the force value of the micro-load is 10% to 20% of the formal detection force; the process of applying the micro-load adopts closed-loop control output, and the control accuracy is ±5% of the target force value.

[0012] This implementation limits the safe range and control precision of the micro-load, ensuring that the mechanical excitation is sufficient to induce observable deformation while ensuring that the excitation is within the linear elastic range and does not trigger safety clamps or other protective devices. Combined with high-precision closed-loop control, it ensures the data quality of sensitivity calculation and the safety of the testing process.

[0013] As one implementation method, the step of cross-validating and fusing the non-contact gap data and the contact gap data to obtain the final gap measurement result includes: when the difference between the non-contact gap data and the contact gap data is within a preset confidence interval, taking the mean or weighted fusion value of the two as the final gap measurement result; when the difference exceeds the preset confidence interval, triggering a re-measurement mechanism or an anomaly marking mechanism.

[0014] This implementation constructs a dual verification logic, using two sensors based on different physical principles as redundant backups for each other; when the data is consistent, the accuracy is improved through fusion; when the data conflict, error correction or alarm is automatically triggered, which can prevent misjudgment caused by the failure of a single sensor or sudden environmental changes, and enhance the fault tolerance of the system.

[0015] As one implementation method, the process of acquiring the non-contact gap data includes: using a visual recognition module to track the changes in pixel coordinates on both sides of the door gap, and converting the pixel displacement into a physical gap change based on a pre-calibrated pixel equivalent; wherein, optical flow, template matching algorithm, or edge tracking algorithm is used to perform the tracking function, and the pixel equivalent is calibrated offline using a checkerboard calibration plate or online using the lower sill markings; the process of acquiring the contact gap data includes: using a capacitive displacement sensor to collect contact gap data.

[0016] This implementation method clarifies the specific implementation path of non-contact measurement, combines a variety of robust visual tracking algorithms and flexible calibration strategies, and adapts to the uneven lighting and complex texture environment in elevator shafts; the use of capacitive displacement sensors ensures the accuracy of contact measurement and provides high-quality basic data for multi-source fusion.

[0017] As one implementation, before applying the formal detection force, the method further includes: converting the pixel coordinates of the most unfavorable force application point into physical coordinates through a pre-calibrated hand-eye relationship matrix, and adjusting the posture of the force application push rod so that its force application direction is perpendicular to the door surface; during the application of the formal detection force, a constant force coil spring mechanism is used to maintain a constant output thrust.

[0018] This implementation method ensures the normal consistency of the force application direction through hand-eye calibration and posture adaptive adjustment, eliminating the component force error introduced by oblique force application; it maintains constant thrust by utilizing the mechanical characteristics of the constant force coil spring mechanism, which has better force value stability and response smoothness compared to pure motor control, ensuring the accuracy of the mechanical conditions for compliance testing.

[0019] As one implementation method, before acquiring the static appearance features of the bottom of the door leaf, the method further includes: acquiring a high-definition image of the bottom area of ​​the door leaf and performing distortion correction; locating the lower sill baseline based on the edge of the lower sill, the lower sill engraving line, or a preset calibration feature; establishing a detection coordinate system with the lower sill baseline and the center line of the door gap as references; under the detection coordinate system, limiting the bottom area of ​​the door leaf to a preset height range above the bottom edge of the door leaf and the area that the force application end can contact; and dividing it into multiple candidate force application areas according to a preset physical size grid, or adaptively dividing it according to the bottom edge contour of the door leaf, the edge of the door gap, and the boundary of the contactable area.

[0020] This implementation establishes a standardized spatial reference system and candidate region division rules. It overcomes imaging and environmental interference through distortion correction and benchmark compensation. It ensures the matching of the physical size of the candidate region and the force application end through gridding or adaptive division, laying a geometric foundation for subsequent feature extraction and force application operations.

[0021] As one implementation method, after obtaining the final gap measurement result, the method further includes: comparing the final gap measurement result with a preset safety threshold; when the final gap measurement result exceeds the preset safety threshold, determining that the door leaf has a risk of being opened, and recording the opening distance, force application position, force application curve, image evidence, and detection time.

[0022] This implementation method enables automated judgment of test results and retention of the entire evidence chain. It not only provides clear safety conclusions but also records complete mechanical and visual raw data, providing detailed and reliable evidence for subsequent fault tracing, maintenance decisions, and remote monitoring.

[0023] The beneficial effects of this invention are as follows: The intelligent detection method for elevator door gap safety provided in this application constructs a two-level optimization system: "static appearance feature coarse screening + micro-load dynamic mechanical response fine screening." It utilizes visual recognition to extract multi-dimensional structural weaknesses at the bottom of the door leaf for preliminary quantitative screening, and then combines this with mechanical verification of the gap change sensitivity under micro-load. This fundamentally solves the technical problem of subjective and inefficient blind trial-and-error selection in traditional manual inspection, achieving objective, accurate, and non-destructive positioning of the most unfavorable force application point. Simultaneously, this application employs a multi-source data fusion mechanism combining non-contact visual measurement and contact displacement sensing. During the formal force application process, both types of gap data are simultaneously acquired and cross-validated. The complementary characteristics of heterogeneous sensors effectively suppress interference from environmental factors such as oil stains, light, and vibration on a single measurement method, significantly improving the robustness and confidence of gap measurement results. Furthermore, the synergistic cooperation between the above two-level optimization system and the multi-source data fusion mechanism forms a complete automated detection chain from candidate area screening and most unfavorable point locking to precise gap measurement, eliminating the randomness introduced by human judgment in traditional inspection and improving the efficiency and consistency of inspection results. Attached Figure Description

[0024] Figure 1 This is a flowchart of an intelligent detection method for elevator landing door gap safety according to an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. Example 1

[0027] like Figure 1 As shown, this embodiment provides an intelligent detection method for elevator landing door gap safety. This method constructs a three-stage detection logic of "static coarse screening + dynamic fine screening + multi-source fusion," aiming to solve the problems of subjective force point selection and easy interference in gap measurement in traditional detection methods. Specifically, the detection method includes the following steps:

[0028] Step S1 involves acquiring the static appearance features of the bottom of the door leaf. Based on these features, multiple candidate force application areas on the bottom of the door leaf are initially screened to obtain preliminary candidate areas. Specifically, static appearance features refer to physical attributes that reflect the structural integrity or potential weak points of the bottom of the door leaf, collected through visual or other sensing means without applying external force. Examples include geometric deformation, surface corrosion, foreign object obstruction, or abnormal edge contours. Candidate force application areas are several discrete test units pre-divided on the bottom of the door leaf. The division method can be a fixed-size grid or an area adaptively generated based on the edge contour of the door leaf. The core of this step is to establish a mapping relationship between "appearance features" and "structural weakness probability," transforming the subjective judgment that originally relied on human experience into a quantitative screening process based on objective data. Through this step, areas that obviously do not meet the testing conditions can be quickly eliminated from numerous candidate areas, locking in several potentially risky preliminary candidate areas. This significantly narrows the search range for subsequent mechanical testing, improves detection efficiency, and avoids unnecessary loading tests on normal areas.

[0029] Step S2 involves applying a micro-load to the initial screening candidate area and obtaining the dynamic mechanical response of the candidate area under the micro-load. Based on the dynamic mechanical response, the most unfavorable force application point is determined from the initial screening candidate area. The force value of the micro-load is less than the formal testing force. In this embodiment, the micro-load is defined as a probing excitation with a force value significantly lower than the formal testing force used for compliance determination, for example, only 10% to 20% or less of the formal testing force, to ensure that plastic deformation of the door structure or triggering of safety protection devices does not occur during the test. The dynamic mechanical response refers to the elastic deformation characteristics of the door under the micro-load, such as the rate of displacement change, the reciprocal of stiffness, or vibration attenuation characteristics. This step constitutes the second-level optimization mechanism of this method, namely "dynamic fine screening." Unlike judging solely by appearance, micro-load testing directly addresses the mechanical essence of the structure, enabling the identification of hidden weak points that have no obvious external defects but have loose internal connections or insufficient support stiffness. By comparing the response intensity of each preliminary candidate area under the same micro-load excitation, the system can objectively determine the point with the most sensitive response (i.e. the lowest stiffness) as the most unfavorable force application point, thereby completely eliminating the randomness and blindness of manually selecting the force application location and ensuring the effectiveness of subsequent formal testing.

[0030] Step S3 involves applying a formal detection force at the most unfavorable point of application, and simultaneously acquiring non-contact and contact gap data during the application process. The formal detection force refers to the rated load specified in relevant safety standards for verifying the door's ability to remain closed, such as 150N. In this step, "synchronous acquisition" is a crucial technical requirement, meaning that the timestamps of the non-contact and contact measurements must be strictly aligned with the force application process to ensure that the collected gap data accurately reflects the door's state under that specific force value. Non-contact gap data typically originates from vision systems, offering advantages such as full-field coverage and no additional load, but is susceptible to lighting and texture effects. Contact gap data, on the other hand, comes from physical probes such as capacitive displacement sensors, offering high accuracy and resistance to light interference, but is a single-point measurement and may introduce minute contact forces. This embodiment, by simultaneously acquiring these two types of heterogeneous data during the force application process, provides spatiotemporally aligned raw materials for subsequent data fusion, ensuring the completeness of the measurement dimensions.

[0031] Step S4 involves cross-validation fusion of non-contact and contact gap data to obtain the final gap measurement result. Cross-validation fusion is not a simple numerical averaging, but a data processing strategy based on confidence assessment. Specifically, when the consistency between the two types of data is high, weighted fusion can further improve measurement accuracy; when significant deviations occur, it indicates that one type of sensor may be affected by environmental interference under the current operating conditions (such as visual failure due to oil obstruction, or abnormal contact sensor readings due to foreign objects in the door gap). In this case, the system can trigger a retest mechanism or an anomaly marker instead of directly outputting an erroneous result. This multi-source redundancy design utilizes the physical complementarity of the two measurement principles, effectively overcoming the limitations of a single measurement method in the complex environment of the elevator shaft, significantly improving the robustness and reliability of the gap measurement results, and providing solid data support for the final safety determination.

[0032] It should be understood that although this embodiment uses an elevator landing door as an example, the above-mentioned detection logic of "static coarse screening + dynamic fine screening + multi-source fusion" is also applicable to other scenarios that need to assess the safety of the force gap of door panels or plate structures, as long as the corresponding functional requirements are met. Example 2

[0033] Building upon Example 1, this example further refines the specific implementation of the preliminary screening based on static appearance features in step S1. As one implementation method, preliminary screening of multiple candidate force application areas at the bottom of the door leaf is performed based on static appearance features to obtain initial candidate areas. This includes: extracting weak feature parameters characterizing the structural weakness of the bottom of the door leaf; standardizing the weak feature parameters and inputting the standardized weak feature parameters into a multi-feature fusion scoring model to obtain the most unfavorable score for each candidate force application area; and selecting a predetermined number of candidate force application areas as initial candidate areas based on the most unfavorable scores from high to low. This process transforms the originally abstract "structural weakness" into a calculable and comparable quantitative indicator, and comprehensively assesses risk by fusing multi-dimensional information through a model, avoiding the one-sidedness of judging by a single feature.

[0034] Specifically, to comprehensively and accurately characterize the structural state of the bottom of the door leaf, weak feature parameters encompass multiple dimensions, from geometric shape to surface texture. In this embodiment, weak feature parameters include: local pixel width of the door gap, continuity of the bottom edge contour of the door leaf, the suspension geometry of the door leaf relative to the lower sill baseline, area of ​​the accessible region, accessibility of the force-applied end, curvature of the bottom edge deformation of the door leaf, probability of foreign objects in the gap area, quantified value of the degree of corrosion at the bottom edge of the door leaf, rate of change of curvature of the bottom edge contour of the door leaf, and confidence level of foreign object detection inside the door gap. These parameters do not exist in isolation but constitute a complementary feature space. For example, the local pixel width and edge deformation curvature of the door gap mainly reflect macroscopic geometric deformation, directly indicating whether the door leaf has bent or misaligned; while the quantified value of the degree of corrosion at the bottom edge of the door leaf and the probability of foreign objects in the gap area focus on microscopic surface degradation. These defects are often not visually obvious, but they can significantly weaken the mechanical properties of the material; the reachability of the force application end and the area of ​​the contactable region are constraints from the perspective of the feasibility of detection execution, ensuring that the selected areas are not only theoretically weak, but can also be effectively force-applied in actual operation. This multi-dimensional parameter combination produces a synergistic effect, enabling the system to identify hidden weak points that are normal in a single indicator but have hidden dangers under the overall condition. For example, although the door gap width in a certain area does not exceed the standard, it is accompanied by slight corrosion and abnormal contour curvature, and its actual risk is far higher than the result of judging by the width alone.

[0035] After obtaining the aforementioned multidimensional heterogeneous parameters, due to the significant differences in the physical dimensions and numerical ranges of each parameter (e.g., pixel widths ranging from tens to hundreds of integers, while the rate of curvature change is a decimal), directly inputting them into the model would lead to features with larger numerical values ​​dominating the training process. Therefore, this embodiment uses the Z-score normalization method to process the weak feature parameters, converting them into standard normal distribution data with a mean of 0 and a standard deviation of 1. This step eliminates the influence of dimensions, making different features comparable in the model and ensuring the fairness and accuracy of subsequent fusion scoring.

[0036] Subsequently, the standardized weak feature parameters are input into the multi-feature fusion scoring model. In this embodiment, the multi-feature fusion scoring model is either a gradient boosting regression tree model or a fully connected neural network model. Both models possess strong nonlinear fitting capabilities, enabling them to capture complex interactions between features rather than simple linear weighting. For example, the gradient boosting regression tree model, by constructing multiple decision trees and iteratively optimizing the residuals, can automatically learn nonlinear rules such as "when the degree of corrosion is high and the rate of curvature change is large, the risk increases exponentially"; the fully connected neural network, through multi-layer neuron transformations, maps high-dimensional features to a unified risk semantic space. The model outputs a worst-case scenario score, which is a normalized continuous value; the higher the score, the greater the likelihood that the candidate application area is the worst-case scenario. Finally, the system sorts the candidates from highest to lowest based on the worst-case scenario scores and selects the top N (N≥3) candidate application areas as initial screening candidate areas. This Top N screening strategy not only retains enough candidate samples for subsequent dynamic fine screening and verification to prevent the omission of the true worst point due to model misjudgment, but also effectively eliminates a large number of low-risk areas, greatly reducing the number of subsequent micro-load tests, and significantly improving the overall detection efficiency while ensuring detection reliability. Example 3

[0037] Based on Example 1, this example further refines the specific implementation of applying a micro-load to the preliminary screening candidate area and determining the most unfavorable force application point based on the dynamic mechanical response in step S2. As one implementation method, applying a micro-load to the preliminary screening candidate area and obtaining the dynamic mechanical response of the preliminary screening candidate area under the micro-load, and determining the most unfavorable force application point from the preliminary screening candidate area based on the dynamic mechanical response, includes: measuring the micro-displacement change of the door gap in real time during the application of the micro-load to the preliminary screening candidate area; using the ratio of the door gap increment to the corresponding force increment as the gap change sensitivity; and selecting the preliminary screening candidate area with the highest gap change sensitivity as the most unfavorable force application point.

[0038] Specifically, this process transforms the abstract "mechanical response" into quantifiable physical indicators. Gap change sensitivity, in its physical essence, characterizes the flexibility of the local structure of the door panel, i.e., the reciprocal of its stiffness. When the applied force increment is the same, a larger increase in the door gap indicates a weaker resistance to deformation in that area, a looser structural connection, or lower material stiffness, thus making it more likely to be the weakest point with the highest safety risk. In actual implementation, the system controls the force application mechanism to linearly load each initial screening candidate area, and synchronously collects force and displacement data at a frequency of no less than 30Hz using force sensors and capacitive displacement sensors configured on the force application mechanism. By calculating the slope of the linear segment of the loading curve, i.e. The sensitivity to gap changes was obtained. ,in To increase the door gap, This corresponds to the applied force increment. The system automatically compares the sensitivity values ​​of all initial screening candidate areas and locks the area corresponding to the maximum sensitivity as the most unfavorable force application point. This objective screening mechanism based on mechanical transmission characteristics effectively compensates for the inability to identify internal connection defects or latent fatigue damage based solely on static appearance features, ensuring that subsequent formal testing can accurately target the real risk points.

[0039] To ensure that the micro-load test effectively induces observable structural deformation without damaging the door or interfering with the normal operation of the elevator, this embodiment strictly limits the range of micro-load force values. As one implementation method, the micro-load force value is 10% to 20% of the formal test force; the application of the micro-load employs closed-loop control output with a control accuracy of ±5% of the target force value. For example, when the formal test force is the standard-specified 150N, the micro-load force value should be controlled between 15N and 30N. In specific implementations, the specific value within this range can be flexibly selected based on the door material, service life, or on-site conditions. For instance, 10% (15N) can be used for old, severely corroded door panels to prevent overload; 15% (22.5N) can be used as the standard excitation for routine testing; or 20% (30N) can be used for new, high-strength door panels with greater stiffness to enhance signal response.

[0040] The numerical range is not arbitrarily chosen, but rather determined as the optimal detection window based on the elastic response characteristics of the door leaf under normal conditions and signal quality requirements. If the micro-load is less than 10% of the formal detection force, the generated micro-displacement signal of the door gap is easily drowned out by background noise such as environmental vibration, sensor noise, or mechanical transmission clearance due to insufficient excitation energy, resulting in a low signal-to-noise ratio, large dispersion of the calculated sensitivity data, and even spurious peaks, thus leading to misjudgment or missed detection. Conversely, if the micro-load is more than 20% of the formal detection force, it may exceed the linear elastic response range, causing the force-displacement relationship to exhibit nonlinear characteristics, distorting the sensitivity index calculated based on the linear slope, and affecting the reliability of sensitivity ranking in different candidate intervals and the repeatability of the test. In addition, for problematic door leaves with serious structural defects, even in the micro-load stage, they will exhibit abnormally high sensitivity to gap changes, which is itself an effective risk warning signal and does not require increasing the preload force for identification. Therefore, the range of 10% to 20% is the optimal balance point between ensuring signal measurability and maintaining linear elastic response, ensuring data quality and non-destructive characteristics in the micro-load detection stage.

[0041] Furthermore, the accuracy of closed-loop control during the application of micro-loads is crucial to the accuracy of sensitivity calculation. Since the sensitivity to gap changes is derived from the ratio of force increment to displacement increment, significant fluctuations or steady-state errors in the applied force will directly distort the ratio calculation, thus affecting the reliability of sensitivity ranking across different candidate intervals. This embodiment employs closed-loop control to achieve precise micro-load output. Closed-loop control refers to the real-time acquisition and feedback of the actual output force value of the applying mechanism to the controller via a high-precision force sensor. The controller compares the feedback value with the target force value to obtain a deviation signal, which is then dynamically adjusted using a PID control algorithm to form a closed control loop of "output → detection → comparison → adjustment → output". Through this closed-loop mechanism, the system can continuously correct output deviations caused by mechanical transmission gaps, friction fluctuations, or load changes, stabilizing the micro-load output accuracy within ±5% of the target value. This ensures the consistency of excitation conditions and the reliability of data in each test, providing a solid underlying data guarantee for the precise locking of the most unfavorable application point. Example 4

[0042] Based on Example 1, this example further refines the specific implementation methods of multi-source data acquisition and cross-validation fusion in steps S3 and S4. As one implementation method, cross-validation fusion is performed based on non-contact gap data and contact gap data to obtain the final gap measurement result. This includes: when the difference between the non-contact gap data and the contact gap data is within a preset confidence interval, the average or weighted fusion value of the two is taken as the final gap measurement result; when the difference exceeds the preset confidence interval, a re-measurement mechanism or an anomaly marking mechanism is triggered. The core of this fusion strategy lies in utilizing the complementary characteristics of two sensors with different physical principles to construct a heterogeneous redundancy verification system, rather than simply superimposing numerical values. Specifically, the preset confidence interval is a dynamic threshold range determined based on the historical error distribution statistics of the two types of sensors under standard operating conditions, for example, it can be set to ±0.5mm or ±1% of the range. When the difference between the two falls within this interval, it indicates that the current measurement environment is relatively ideal, and both types of data have high reliability. At this time, although a simple arithmetic average can smooth out random noise to some extent, it cannot cope with the situation where the performance of a certain sensor temporarily declines. Therefore, this embodiment preferably employs a weighted fusion method, dynamically assigning weights based on the real-time signal-to-noise ratio (SNR) or signal quality index of each sensor. For example, when a sudden change in illumination within the wellbore causes a decrease in visual image contrast, the system automatically calculates the decrease in the SNR of the visual data, thereby reducing the weight of non-contact gap data while increasing the weight of contact gap data. This makes the final fusion result more favorable to the more reliable sensor source. Compared to a fixed weighted average, this adaptive weighting mechanism significantly improves the system's measurement robustness under dynamic interference environments.

[0043] When the difference between non-contact gap data and contact gap data exceeds a preset confidence interval, it indicates that at least one type of sensor has experienced a significant deviation or malfunction. In this case, the system will not directly output a potentially erroneous measurement value, but will instead trigger a re-measurement mechanism or an anomaly marking mechanism. The re-measurement mechanism may include unloading and reloading the force application mechanism, adjusting visual exposure parameters and re-acquiring images, or performing zero-point calibration on the capacitive grating sensor, and performing a fusion judgment again after the re-measurement. If the difference still exceeds the limit after the re-measurement, the anomaly marking mechanism is activated, recording the current raw data, the magnitude of the difference, environmental parameters, and sensor status code, and marking the detection result as "awaiting manual verification" to prevent safety risks caused by automated misjudgment. This hierarchical response mechanism ensures efficient automatic detection under normal operating conditions while retaining error correction opportunities and traceability basis for abnormal operating conditions, reflecting the completeness and safety of the system design.

[0044] To support high-quality data input for the aforementioned fusion strategy, this embodiment also specifically optimizes the acquisition process for the two types of gap data. As one implementation method, the process of acquiring non-contact gap data includes: using a visual recognition module to track the changes in pixel coordinates on both sides of the door gap, and converting the pixel displacement into a physical gap change based on a pre-calibrated pixel equivalent; wherein, optical flow, template matching, or edge tracking algorithms are used to perform the tracking function, and the pixel equivalent is calibrated offline using a checkerboard calibration board or online using the lower sill markings. In the special scenario of elevator shafts, door surfaces often suffer from texture loss, reflection, or oil stains, and lighting conditions fluctuate with changes in the car's position. To address these challenges, optical flow utilizes the spatiotemporal continuity of pixel intensity in an image sequence to estimate the motion field, maintaining good tracking stability even with blurred edges; template matching algorithms establish multi-dimensional feature templates for the door gap feature regions and perform similarity searches in subsequent frames, exhibiting strong robustness to lighting changes; edge tracking algorithms combine Canny or Sobel operators to extract sub-pixel-level edges, suitable for scenarios with clear textures but low contrast. The system can adaptively switch or combine the above algorithms based on the real-time image quality assessment results to ensure that continuous and reliable pixel displacement data can be obtained under various working conditions.

[0045] Regarding pixel equivalent calibration, while traditional offline calibration offers high accuracy, it struggles to compensate for parameter drift caused by minor deformations of the camera mounting bracket, lens thermal expansion and contraction, or long-term vibration. Therefore, this embodiment introduces an online calibration mechanism utilizing the sill markings. Elevator landing sills typically have markings or reference grooves with known spacing, and these physical features can be captured by the vision recognition module during each inspection. The system calculates the pixel distance between adjacent markings in the image in real time, deduces the current actual pixel equivalent, and compares it with the offline calibration value for correction. This online self-calibration capability effectively eliminates the accumulation of systematic errors during long-term operation, ensuring consistent accuracy of non-contact measurements throughout their entire lifecycle. Simultaneously, the process of acquiring contact gap data includes using a capacitive grating displacement sensor. Based on the capacitive grid coupling principle, capacitive grating displacement sensors offer advantages such as high resolution, fast response speed, and strong resistance to oil and moisture interference, making them particularly suitable for harsh environments prone to dust and oil accumulation, such as elevator pits or landing door bottoms. The absolute displacement signal output by the sensor and the non-contact visual data form a positive interaction at the physical level: vision excels at full-field observation but is sensitive to surface conditions, while capacitive grating excels at precise single-point measurement but relies on physical contact. Their collaborative work provides a solid and reliable data foundation for the aforementioned cross-validation fusion, jointly ensuring the accuracy and environmental adaptability of the final gap measurement results. Example 5

[0046] Beyond the core detection logic described in the preceding embodiments, this embodiment further details the safeguards and result output mechanisms throughout the entire detection process to ensure reproducibility, mechanical accuracy, and traceability in practical engineering applications. These safeguards constitute the necessary support for the technical solution to move from theoretical algorithms to field implementation.

[0047] As one implementation method, before acquiring the static appearance features of the bottom of the door leaf, the method further includes: acquiring a high-resolution image of the bottom area of ​​the door leaf and performing distortion correction; locating the lower sill baseline based on the edge of the lower sill, the lower sill engraving line, or preset calibration features; and establishing a detection coordinate system with the lower sill baseline and the center line of the door gap as references. Within the detection coordinate system, the bottom area of ​​the door leaf is limited to a preset height range above the bottom edge of the door leaf and the area that the force-applying end can contact, and is divided into multiple candidate force-applying areas according to a preset physical size grid, or adaptively divided based on the bottom edge contour of the door leaf, the edge of the door gap, and the boundary of the contactable area. Specifically, since industrial cameras typically use wide-angle lenses to cover the entire bottom field of view of the door leaf, the original image inevitably has barrel distortion. If not corrected, the extracted geometric features such as the door gap width and edge curvature will have systematic deviations, directly affecting the accuracy of subsequent initial screening and scoring. Therefore, this step first uses the camera intrinsic parameter matrix to perform distortion correction on the high-resolution image to restore the true physical spatial proportions. Based on this, locating the lower sill baseline is crucial for establishing a unified spatial reference system. Under ideal conditions, the metal edge of the lower sill can be directly identified as a reference; however, in actual elevator shafts, the lower sill edge is often covered by oil or blurred due to long-term wear. To address this, the system employs a robust linear fitting algorithm (such as RANSAC) to extract unobstructed edge points, or utilizes pre-processed artificial features such as engraved lines and reflective markings on the lower sill for auxiliary positioning. Even in extreme obstruction situations, historical calibration parameters are invoked for compensation, ensuring the detection coordinate system remains stable and reliable under various field conditions. Regarding the division of candidate force application areas, the preset physical size grid division is suitable for standardized scenarios with flat door bottoms and regular structures. For example, the effective test area is uniformly divided into a 10mm×10mm square matrix for easy traversal. Adaptive division, on the other hand, is designed for old, deformed, or irregularly shaped door panels. The system dynamically generates candidate areas with irregular shapes that conform to the actual structure based on the real-time identified bottom edge contour of the door panel, the direction of the door gap, and the reachability boundary of the force application end. This adaptive strategy avoids the problem of invalid partitions or omissions of key weak areas at the edges of the fixed mesh, ensuring that the candidate area set not only fully covers potential risk points, but also matches the physical contact area of ​​the force-applying push rod, providing a geometrically feasible operational basis for subsequent steps.

[0048] As one implementation method, before applying the formal detection force, the method further includes: converting the pixel coordinates of the most unfavorable force application point into physical coordinates using a pre-calibrated hand-eye relationship matrix, and adjusting the posture of the force-applying push rod so that its force direction is perpendicular to the door surface; during the application of the formal detection force, a constant force coil spring mechanism is used to maintain a constant output push force. This step solves the mapping accuracy problem between the visual recognition space and the mechanical execution space, as well as the mechanical stability problem during the force application process. Specifically, the most unfavorable force application point output by the vision module is located in the image pixel coordinate system, while the force application mechanism operates in its own mechanical coordinate system. There is a non-linear transformation relationship between the two due to factors such as installation position, angle deviation, and mechanical tolerance. This embodiment uses a pre-calibrated hand-eye relationship matrix (usually a 4×4 homogeneous transformation matrix) to accurately convert the pixel coordinates into physical three-dimensional coordinates that the force application mechanism can execute, eliminating positioning errors introduced by camera installation tilt or bracket micro-deformation. More importantly, the door gap detection standard requires that the force applied must be perpendicular to the door surface. Applying force at an angle will generate a component force parallel to the door surface, which will not only lead to an underestimation of the gap measurement value, but may also cause the door to slide or get stuck, resulting in misjudgment. Therefore, after the system moves to the target position, it will adjust the posture of the force-applying push rod in real time according to the normal angle of the door surface to ensure that the push force vector acts strictly along the normal direction. In terms of force output, although the servo motor can control the thrust through the current loop, its output is easily affected by the backlash of the reducer, friction fluctuations, and power supply ripple, making it difficult to maintain an absolutely constant static thrust for a long time. This embodiment introduces a constant force coil spring mechanism as a mechanical buffer and steady-state maintenance element. The constant force coil spring utilizes the elastic potential energy release characteristics of a special alloy strip to provide a nearly constant tension or thrust within its working stroke. When the drive motor compresses or stretches the coil spring to the preset working point, the mechanical characteristics of the coil spring automatically take over the force output, effectively filtering out high-frequency torque pulsations in the motor drive chain, so that the formal detection force applied to the door remains highly stable throughout the entire measurement cycle. This electromechanical co-operational constant force control method has better anti-interference capabilities and long-term stability compared to pure electric control schemes, ensuring that strict compliance testing conditions are met.

[0049] As one implementation method, after obtaining the final gap measurement result, the method further includes: comparing the final gap measurement result with a preset safety threshold; when the final gap measurement result exceeds the preset safety threshold, it is determined that the door leaf has an opening risk, and the opening distance, force application position, force application curve, image evidence, and detection time are recorded. This step realizes automated judgment of the detection result and full-process evidence chain retention, which is a key link in improving the standardization level and traceability of the detection. Specifically, the preset safety threshold is set according to national or industry elevator safety technical specifications. For example, for passenger elevator landing doors, this threshold is usually 6mm or 10mm (depending on the specific standard version). The system automatically compares the fused final gap measurement result with this threshold, eliminating the need for manual reading and judgment, and completely eliminating the risk of subjective misreading. Once an opening risk is determined, the system not only outputs a "non-compliant" conclusion, but also simultaneously generates a complete evidence package containing multi-dimensional information. Among these features, the opening distance, i.e., the final gap measurement, provides a quantified level of risk; the force application location is recorded in the form of physical coordinates and corresponding local images of the door leaf, clarifying the spatial distribution of risk points; the force application curve records the force-displacement relationship throughout the entire process from the initial application of micro-load to the formal testing force, which can be used for post-event analysis to determine whether the mechanical response characteristics of the door leaf are abnormal; image evidence includes close-ups of the door gap and panoramic photos at the moment of force application, intuitively showing the door leaf status and environmental conditions at that time; and the testing time is accurate to the second-level timestamp, linked to the elevator operation log and maintenance records. This evidence chain mechanism not only provides clear rectification guidelines for on-site maintenance personnel, but also provides tamper-proof data support for remote spot checks by regulatory authorities, accident liability determination, and historical trend analysis, making each test result legally traceable and promoting the transformation and upgrading of elevator safety testing from experience-driven to data-driven.

[0050] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent detection method for elevator landing door gap safety, characterized in that, include: S1. Obtain the static appearance features of the bottom of the door leaf, and perform preliminary screening of multiple candidate force application areas at the bottom of the door leaf based on the static appearance features to obtain preliminary candidate areas; S2. Apply a micro-load to the preliminary screening candidate area and obtain the dynamic mechanical response of the preliminary screening candidate area under the action of the micro-load. Based on the dynamic mechanical response, determine the most unfavorable force application point from the preliminary screening candidate area; wherein, the force value of the micro-load is less than the formal detection force; S3. Apply the formal detection force at the most unfavorable force application point, and simultaneously acquire non-contact gap data and contact gap data during the force application process; S4. Based on the non-contact gap data and the contact gap data, cross-validation and fusion are performed to obtain the final gap measurement result; In step S1, based on the static appearance features, multiple candidate force application areas at the bottom of the door leaf are initially screened to obtain preliminary candidate areas, including: S11. Extract weak feature parameters that characterize the weakness of the bottom structure of the door leaf; S12. Standardize the weak feature parameters and input the standardized weak feature parameters into the multi-feature fusion scoring model to obtain the most unfavorable score of each candidate force application region; S13. Select a preset number of candidate force application areas as the initial screening candidate areas based on the worst-case scenario score from high to low.

2. The intelligent detection method for elevator landing door gap safety according to claim 1, characterized in that, The weak feature parameters include: local pixel width of the door gap, continuity of the bottom edge contour of the door leaf, suspension geometry of the door leaf relative to the lower sill baseline, area of ​​the accessible region, reachability of the force application end, deformation curvature of the bottom edge of the door leaf, probability of foreign objects in the gap area, quantification value of the degree of corrosion of the bottom edge of the door leaf, rate of change of curvature of the bottom edge contour of the door leaf, and confidence level of foreign object detection inside the door gap. The multi-feature fusion scoring model is either a gradient boosting regression tree model or a fully connected neural network model.

3. The intelligent detection method for elevator landing door gap safety according to claim 1, characterized in that, In step S2, a micro-load is applied to the initial screening candidate region, and the dynamic mechanical response of the initial screening candidate region under the micro-load is obtained. Based on the dynamic mechanical response, the most unfavorable force application point is determined from the initial screening candidate region, including: S21. During the process of applying the micro-load to the initial screening candidate area, the micro-displacement change of the door gap is measured in real time; S22. The ratio of the increase in door gap to the corresponding increase in applied force is used as the sensitivity to gap change; S23. Select the candidate region with the highest sensitivity to gap changes as the most unfavorable stress point.

4. The intelligent detection method for elevator landing door gap safety according to claim 3, characterized in that, The force value of the micro-load is 10% to 20% of the formal detection force.

5. The intelligent detection method for elevator landing door gap safety according to claim 1, characterized in that, In step S4, cross-validation and fusion are performed based on the non-contact gap data and the contact gap data to obtain the final gap measurement result, including: S41. When the difference between the non-contact gap data and the contact gap data is within a preset confidence interval, the mean or weighted fusion value of the two is taken as the final gap measurement result; S42. When the difference exceeds the preset confidence interval, a remeasurement mechanism or an anomaly marking mechanism is triggered.

6. The intelligent detection method for elevator landing door gap safety according to claim 5, characterized in that, The process of acquiring the non-contact gap data includes: using a visual recognition module to track the changes in pixel coordinates on both sides of the door gap, and converting the pixel displacement into a physical gap change based on a pre-calibrated pixel equivalent; wherein, optical flow, template matching algorithm or edge tracking algorithm is used to perform the tracking function, and the pixel equivalent is calibrated offline using a checkerboard calibration plate or online using the lower sill markings; The process of acquiring the contact gap data includes: using a capacitive displacement sensor to collect the contact gap data.

7. The intelligent detection method for elevator landing door gap safety according to claim 1, characterized in that, Before S3, it also includes: The pixel coordinates of the most unfavorable force application point are converted into physical coordinates by a pre-calibrated hand-eye relationship matrix, and the posture of the force application push rod is adjusted so that the force application direction is perpendicular to the door surface. During the application of the formal detection force, a constant force coil spring mechanism is used to maintain a constant output thrust.

8. The intelligent detection method for elevator landing door gap safety according to claim 1, characterized in that, Before S1, it also includes: High-definition images of the bottom area of ​​the door leaf are acquired and distortion correction is performed. The lower sill baseline is located based on the edge of the lower sill, the lower sill engraving line, or the preset calibration features. A detection coordinate system is established with the lower sill baseline and the center line of the door gap as references. In the detection coordinate system, the bottom area of ​​the door leaf is limited to the area within a preset height range above the bottom edge of the door leaf and the area that the force application end can contact. It is divided into multiple candidate force application areas according to a preset physical size grid, or adaptively divided according to the bottom edge contour of the door leaf, the edge of the door gap and the boundary of the contactable area.

9. The intelligent detection method for elevator landing door gap safety according to claim 1, characterized in that, After obtaining the final gap measurement result in step S4, the following steps are also included: S51. Compare the final gap measurement result with a preset safety threshold; S52. When the final gap measurement result exceeds the preset safety threshold, it is determined that the door leaf has a risk of being opened, and the opening distance, force application position, force application curve, image evidence and detection time are recorded.

Images