A multi-modal self-calibration method for elevator weighing early warning

Abstract

The present application relates to elevator safety control technical field, provide a kind of elevator weighing early warning multi-modal self-calibration method, comprising: by pulling rope type sensor obtains the zero load height value when elevator is in empty load stop state and the full load height value when in full load state;Obtain the zero load height mean of multiple zero load height values in preset time period, judge whether the aging condition of car bottom rubber exists according to the offset of zero load height mean and zero load height value;If the aging condition of car bottom rubber exists, then the latest zero load height mean is updated to old zero load height value, and the full load height value is compensated and calibrated according to the offset, and the latest full load height value is obtained.The present application can effectively reduce the measurement error accumulation caused by long-term use, improve the long-term stability and adaptability of system, not only reduce maintenance cost, but also avoid the safety hazards of falling and tool sliding of operating personnel in shaft operation.

Description

Technical Field

[0001] This invention relates to the field of elevator safety control technology, and more specifically, to a multimodal self-calibration method for elevator weighing early warning. Background Technology

[0002] Traditional elevator weighing systems rely heavily on mechanical pressure sensors at the bottom of the car for load detection. This requires frequent elevator stops and personnel to enter the hoistway or confined space on the car top for calibration and maintenance. Such mechanically-based weighing systems are susceptible to mechanical creep, temperature drift, and uneven wire rope tension. Limited by sensor accuracy, installation location, and environmental interference, they suffer from error accumulation and zero-point drift, making it difficult to achieve long-term, stable, and highly accurate weighing warnings.

[0003] Therefore, it is necessary to develop a multimodal self-calibration method for elevator weighing warning to achieve long-term stable and high-precision elevator weighing warning and reduce maintenance costs. Summary of the Invention

[0004] Therefore, in order to solve the problem of difficulty in achieving long-term stable high-precision weighing warning in traditional elevator weighing systems, this invention provides a multimodal self-calibration method for elevator weighing warning, the specific technical solution of which is as follows: A multimodal self-calibration method for elevator weighing warning includes the following steps: A pull-rope sensor is installed between the car frame and the bottom of the car to obtain the zero-load height value when the elevator is in an unloaded and stopped state and the full-load height value when it is in a fully loaded state. Obtain the average zero-load height of multiple zero-load height values ​​within a preset time period, and determine whether the car floor rubber has aged based on the offset between the average zero-load height and the zero-load height values; If the car floor rubber is aged, the old zero-load height value is updated with the latest zero-load height average value, and the full-load height value is compensated and calibrated according to the offset to obtain the latest full-load height value.

[0005] Compared to traditional manual or periodically calibrated elevator weighing warning systems, the multimodal self-calibration method for elevator weighing warning employs a self-calibration mechanism. This mechanism determines whether the car floor rubber is aging based on the offset between the average zero-load height and the actual zero-load height. If aging is found, the old zero-load height value is updated with the latest average zero-load height, and the full-load height value is compensated for based on the offset to obtain the latest full-load height value. This compensates for errors caused by environmental factors. This mechanism effectively reduces the accumulation of measurement errors due to long-term use, improving the system's long-term stability and adaptability. It eliminates the need for frequent elevator shutdowns and personnel entering the shaft or confined space above the car for calibration and maintenance, reducing maintenance costs and avoiding safety hazards such as falls and tool slippage for workers operating in the shaft.

[0006] Preferably, the multimodal self-calibration method further includes the following steps: The real-time load height value of the elevator is obtained through the pull-rope sensor; The real-time height difference is obtained based on the real-time load height value and the zero load height value; The real-time load of the car is obtained based on the real-time height difference.

[0007] Preferably, the multimodal self-calibration method further includes the following steps: Determine whether the real-time load height value is less than the latest full load height value. If so, determine that the car is overloaded and activate the elevator's audible and visual alarm.

[0008] Preferably, the specific method for obtaining the real-time load of the car based on the real-time height difference includes the following steps: A polynomial model is constructed based on the car load and the real-time height difference. The polynomial model is fitted, and the real-time load of the car is obtained based on the fitted polynomial model.

[0009] Preferably, the polynomial model is represented as ; in, Indicates the car load. Indicates the real-time height difference. These represent the coefficients of the quartic, cubic, quadratic, and linear terms, as well as a preset constant.

[0010] Preferably, according to the formula Compensation calibration is performed on the full load height value; in, Indicates the offset. These represent the zero-load height value and the average zero-load height, respectively. This indicates the full-load height value before compensation calibration. This indicates the full-load height value after compensation calibration. This represents the compensation coefficient.

[0011] Preferably, the multimodal self-calibration method further includes the following steps: The system acquires real-time image data of the elevator car's interior, and uses this data to determine the number of passengers and whether any electric vehicles have entered the car. If the number of passengers exceeds a preset threshold or an electric vehicle enters the elevator car, the elevator's audible and visual alarm will be activated.

[0012] Preferably, the body of the pull-rope sensor is fixedly installed on the central rigid support point at the bottom of the car, and the pull-rope axis of the pull-rope sensor is set in the vertical direction and the stretching direction of the pull-rope is consistent with the sinking direction of the car.

[0013] Preferably, real-time image data is acquired by a network camera installed at a diagonal position on the top of the car away from the car door, and the number of passengers is obtained by identifying passengers based on the YOLO object detection algorithm.

[0014] Preferably, when the number of passengers detected is zero and the data fed back by the pull-cord sensor is stable and remains at a preset time value, the elevator is determined to be in an unloaded and stopped state. Attached Figure Description

[0015] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0016] Figure 1 This is a schematic diagram of the overall process of a multimodal self-calibration method for elevator weighing early warning in one embodiment of the present invention; Figure 2 This is a flowchart illustrating a multimodal self-calibration method for elevator weighing early warning, as described in another embodiment of the present invention. Figure 1 ; Figure 3 This is a flowchart illustrating a specific method for obtaining the real-time load of the car in one embodiment of the present invention; Figure 4 This is a flowchart illustrating a multimodal self-calibration method for elevator weighing early warning, as described in another embodiment of the present invention. Figure 2 ; Figure 5 This is a schematic diagram of the overall workflow of the system in one embodiment of the present invention; Figure 6This is a schematic diagram of the curve relationship between the car load and the raw data of the rope sensor in one embodiment of the present invention; Figure 7 This is a schematic diagram of the curve relationship between car load and relative height in one embodiment of the present invention; Figure 8 This is a schematic diagram of the debugging software interface in one embodiment of the present invention; Figure 9 This is a schematic diagram of the recognition effect of the visual recognition module in one embodiment of the present invention; Figure 10 This is a schematic diagram of the encoding format for reading registers and sending back encoding data of a pull-string sensor encoder in one embodiment of the present invention; Figure 11 This is a schematic diagram of the hardware deployment of each module in one embodiment of the present invention; Figure 12 This is a flowchart of the self-calibration function program in one embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.

[0018] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0019] 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 invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0020] In this invention, "first" and "second" do not represent a specific quantity or order, but are merely used to distinguish names.

[0021] Before describing the embodiments of the present invention, we will first briefly introduce the prior art.

[0022] Traditional elevator weighing technology has many pain points and shortcomings, mainly manifested in the following aspects: 1. The weighing accuracy is insufficient, it is affected by environmental factors, the long-term stability of the sensor is poor, and it is difficult to accurately monitor the elevator load. 2. Traditional sensors are susceptible to aging of the car floor adhesive, requiring frequent manual calibration, resulting in high maintenance costs and low efficiency; 3. There is a general lack of effective real-time monitoring methods for passenger numbers, which poses a safety hazard due to overloading without timely alarms; 4. To ensure safety, maintenance personnel often set the overload alarm threshold too low, reducing the elevator's load capacity and resulting in a waste of resources and energy.

[0023] 5. The elevator relies heavily on mechanical pressure sensors at the bottom of the car for load detection, which requires frequent elevator stops and the dispatch of personnel into the confined space of the shaft or car top for calibration and maintenance. Manual calibration takes 2-3 hours per session, which increases the average annual maintenance cost and poses risks of falls and tool slippage during shaft operations.

[0024] like Figure 1 As shown, in order to solve the problem of difficulty in achieving long-term stable high-precision weighing warning in traditional elevator weighing systems, a multimodal self-calibration method for elevator weighing warning in one embodiment of the present invention includes the following steps: S1, A pull-rope sensor is installed between the car frame and the bottom of the car to obtain the zero-load height value when the elevator is in an unloaded and stopped state and the full-load height value when it is in a fully loaded state.

[0025] Preferably, the method for obtaining the elevator's empty and stopped state includes: acquiring real-time image data through a network camera installed on the top of the car at a diagonal position away from the car door, and identifying passengers based on the YOLO object detection algorithm to obtain the number of passengers.

[0026] Specifically, such as Figure 11 As shown, this embodiment uses a development board as the controller for the elevator system. The network camera is fixedly installed at a downward angle of 25 degrees or 30 degrees, covering the entire interior area of ​​the car, and avoiding direct contact with the car lights to prevent overexposure of the image due to strong light. The corresponding data and signal cables are routed through pre-drilled holes in the car top and connected to the embedded development board and the car top power supply, respectively. The USB 3.0 interface of the development board is connected to the pull-cord sensor through a signal conversion module (specifically, an RS485-USB conversion module: CH340 chip, bidirectional half-duplex) to realize communication between the pull-cord sensor data and the computing control module.

[0027] The pull-rope sensor body is fixedly installed on the central rigid support point at the bottom of the car. Its pull-rope axis is vertically oriented, and the tension direction of the pull-rope is consistent with the downward direction of the car. The pull-rope sensor is a high-precision pull-rope displacement sensor, model BRT27, with a range of 0-650mm, linear accuracy ±0.05%, protection rating IP67, baud rate of 9600-115200bps, and supports the Modbus-RTU protocol. It is used for real-time monitoring of the vertical displacement of the car bottom caused by load changes.

[0028] The development board uses a LubanCat2 RK3568 card computer (4-core Cortex-A55, 4GB LPDDR4), equipped with a Linux operating system, and has pre-programmed code for pull-wire sensor communication and relay control. It can receive and process data from pull-wire sensors and network cameras and control relay actions. The relay uses an optocoupler-isolated relay module to control the on / off state of the overload alarm signal in the car top control cabinet.

[0029] When the number of passengers detected is zero and the data fed back by the pull-cord sensor remains stable for a preset time value, the elevator is determined to be in an unloaded and stopped state. For example, during normal elevator operation, if the number of passengers N=0 and the height data fed back by the pull-cord sensor remains stable for 10 minutes, the elevator is considered to be in an unloaded and stopped state.

[0030] The YOLO target detection algorithm is deployed on a development board. The development board's gigabit Ethernet port communicates with the network camera via a PoE cable to acquire video stream data from inside the elevator, and then obtains the real-time image data based on the video stream data. A visual interface can be developed on the QT platform, and the complete program can be burned to the development board to achieve local and remote operation. Users can access the QT interface via mobile phone or tablet to view the elevator's operating status in real time and perform remote debugging and management.

[0031] The YOLO object detection algorithm is based on a deep convolutional neural network and is trained by collecting and labeling a large number of datasets from similar application scenarios to optimize detection accuracy. After being trained on a computing server, it is applied to real-time inference of video streams inside elevators. The YOLO object detection algorithm is adapted to the computing resources on the development board and employs model quantization or pruning optimization strategies to improve inference speed and reduce computational costs. The YOLO object detection algorithm model runs on the NPU of the development board and performs inference on the acquired video stream to identify and count the people in the elevator. The detection results are then transmitted to the data processing module to determine whether the elevator is in an empty and stopped state.

[0032] S2, obtain the average zero-load height of multiple zero-load height values ​​within a preset time period, and determine whether the car bottom rubber is aging based on the offset between the average zero-load height and the zero-load height value.

[0033] S3, if the car bottom rubber is aged, the old zero-load height value is updated with the latest zero-load height average value, and the full-load height value is compensated and calibrated according to the offset to obtain the latest full-load height value.

[0034] Whenever the elevator is in an unloaded, stopped state, the system's calculation and control module will use the data from the pull-wire sensor at that time. (i=1, 2, 3...) Records are saved, and the average zero-load height can be calculated weekly. If the average height under zero load and zero load height value If the offset is large (e.g., the offset equivalent load exceeds 50kg), the calculation and control module determines that the car floor rubber is aging, and the car's full-load height needs to be recalibrated. The old zero-load height value is updated using the latest average zero-load height, and the average zero-load height is then updated. and zero load height value The difference between the two (i.e., the offset) multiplied by a preset compensation coefficient The full-load height value is compensated and calibrated to form a new zero-load height value. and full load height value That is, according to the formula The full-load height value is compensated and calibrated. Among other things, Indicates the offset. This indicates the full-load height value before compensation calibration. This indicates the full load height value after compensation calibration.

[0035] New zero-load height value . This represents the sampled data from the pull-wire sensor. The quantity. Then, clear the recorded data. The new zero-load height value and full-load height value will be used in the daily judgment of the car's full load status to achieve automatic compensation for the deviation of the elevator's full and overload alarm value.

[0036] Compared to traditional manual or periodically calibrated elevator weighing warning systems, the multimodal self-calibration method for elevator weighing warning employs a self-calibration mechanism. This mechanism determines whether the car floor rubber is aging based on the offset between the average zero-load height and the actual zero-load height. If aging is found, the old zero-load height value is updated with the latest average zero-load height, and the full-load height value is compensated for based on the offset to obtain the latest full-load height value. This compensates for errors caused by environmental factors. This mechanism effectively reduces the accumulation of measurement errors due to long-term use, improving the system's long-term stability and adaptability. It eliminates the need for frequent elevator shutdowns and personnel entering the shaft or confined space above the car for calibration and maintenance, reducing maintenance costs and avoiding safety hazards such as falls and tool slippage for workers operating in the shaft.

[0037] As a preferred technical solution, such as Figure 2 As shown, the multimodal self-calibration method further includes the following steps: S4, the real-time load height value of the elevator is obtained through the pull-rope sensor; S5, obtain the real-time height difference based on the real-time load height value and the zero load height value; S6, Obtain the real-time load of the car based on the real-time height difference.

[0038] like Figure 3 As shown, in step S6, the specific method for obtaining the real-time load of the car based on the real-time height difference includes the following steps: S61, construct a polynomial model based on the car load and real-time height difference; S62, the polynomial model is fitted, and the real-time load of the car is obtained based on the fitted polynomial model. Here, the polynomial model can be fitted based on the least squares method or the Newton-Raphson iteration method.

[0039] Specifically, the polynomial model is represented as ;in, Indicates the car load. Indicates the real-time height difference. These represent the coefficients of the quartic, cubic, quadratic, and linear terms, as well as a preset constant.

[0040] like Figure 10As shown, the pull-string sensor has a built-in encoder and registers and follows the RS485 communication protocol. After the host computer (such as a development board) completes the serial port configuration (baud rate 9600bps, 8 data bits, no parity, 1 stop bit), it can send standard MODBUS-RTU encoding to the encoder to read and write the registers at the corresponding addresses, thus realizing serial communication with the pull-string sensor. The most important part is reading the distance data between the rope end and the base (i.e., the encoder value). This data is stored in registers with addresses 0x0000~0x0001. The host computer sends the hexadecimal code 01 03 00 00 00 02 C4 0B, and the encoder will return 01 03 04 XX XX XX XX CC 40, where bits 4-7 are the hexadecimal encoder value.

[0041] Write a program to control the host computer to send a command code to the encoder of the pull-string sensor every 0.5 seconds to read the encoder value. After waiting for the complete 9-bit code to be returned, extract bits 4-7 and convert them into a decimal encoder value. Then, calculate the length of the rope end pulled out according to the distance calculation formula, as follows:

[0042] in, X is the current encoder value, and X is the preset reference position encoder value (if the position when the rope end is not pulled out is set as the reference position, then X=0). A is the circumference of the grating encoder wheel (C=60 for BRT27 draw-wire sensor), and A is the single-turn resolution of the grating encoder (A=4096 for BRT27 draw-wire sensor). It is the current distance (in mm) between the sensor rope head and the reference position.

[0043] After the pull-cord sensor and other hardware are correctly installed, the sensor encoder data when the car is unloaded is recorded as follows: (i.e., zero-load height value), the real-time data of the sensor encoder is recorded as Real-time height difference The load-relative height polynomial model fitted during the input data acquisition phase is used to iteratively approximate and solve the fourth-order polynomial with 500 as the guess value, so as to obtain the car load estimate.

[0044] Traditional elevator overload detection systems are based on the mechanical deformation principle of the car bottom damping rubber: the bottom of the elevator car and the car frame are flexibly connected by the car bottom damping rubber. When the car load increases, the car bottom rubber undergoes vertical compression deformation, resulting in a reduction in the distance between the bottom of the car and the car frame. By installing a microswitch on the car frame and using a standard weight to calibrate the switch trigger position, when the load reaches the threshold, the deformation of the car bottom rubber causes the bottom of the car to contact the microswitch, disconnecting the normally closed contact of the elevator main control overload signal, thereby realizing the elevator overload alarm function.

[0045] As elevators age, the rubber sealant on the car floor ages, causing greater deformation under the same load. This can trigger the microswitch before the car load reaches the overload threshold, leading to premature overload alarms and significantly reduced elevator efficiency. Therefore, elevator maintenance companies need to perform regular calibration and maintenance. Current maintenance procedures require loading weights into the car to full load and then recalibrating the microswitch position. This calibration process is cumbersome, disrupts normal elevator operation, and consumes considerable manpower and resources.

[0046] This invention creatively retains the deformation detection principle of the car floor rubber, using a pull-string sensor to replace the original microswitch for distance monitoring. An embedded system continuously monitors sensor data under no-load and no-load stop conditions, and a multi-cycle average comparison algorithm automatically identifies the aging of the car floor rubber, dynamically compensating and updating the full-load threshold. In addition to inheriting the original overload protection function, it also adds load estimation and automatic compensation calibration functions for overload alarm offset.

[0047] As a preferred technical solution, such as Figure 4 As shown, the multimodal self-calibration method further includes the following steps: S7. Determine whether the real-time load height value is less than the latest full load height value. If so, determine that the car is overloaded and activate the elevator's audible and visual alarm.

[0048] S8, acquire real-time image data of the elevator car interior, and obtain the number of passengers and determine whether an electric vehicle has entered the elevator car based on the real-time image data.

[0049] S9. If the number of passengers exceeds a preset threshold or an electric vehicle enters the elevator car, the elevator's audible and visual alarm will be activated.

[0050] Specifically, this embodiment uses YOLOv8 object detection technology to monitor the situation inside an elevator car, including passenger counting and electric vehicle detection. First, video data of different scenes inside the elevator is collected and labeled. A YOLOv8 model is then trained on a computing server, and parameters are optimized to improve detection accuracy. The trained YOLOv8 model is converted to ONNX (Open Neural Network Exchange) format and optimized using RKNN (Rockchip Neural Network). Finally, it is deployed on an RK3568 development board to improve inference efficiency.

[0051] To achieve efficient elevator car monitoring, Hikvision network cameras can be used to acquire RTSP video streams, and the data can be processed through a QT front-end development. OpenCV is used to decode the video frames, and then the RKNN API is called to perform target detection. The recognition results are then displayed in real time on the QT interface.

[0052] To improve the system's adaptability and detection accuracy, the data acquisition phase included elevator car video samples under different times and lighting conditions. Data augmentation techniques were used to expand the dataset, and transfer learning was combined to optimize the model training process. After training, the YOLOv8 model was converted to ONNX format and optimized using the RKNN toolchain, enabling it to achieve faster inference speed and lower computational resource consumption when running on embedded devices.

[0053] By determining whether the elevator car is overloaded, whether the number of passengers exceeds a preset threshold, and whether any electric vehicles have entered the car, and by taking corresponding audible and visual alarm actions, the system can monitor the situation inside the elevator car in real time, improve evacuation efficiency, and reduce safety hazards.

[0054] like Figure 5 As shown, one embodiment of the present invention also provides the testing steps and principles of the development board, as detailed below: like Figure 8 As shown, after completing all hardware installation and wiring, connect the development board and computer to the same local area network. You can then access the development board's operating system and perform control operations via the computer. After opening the pre-programmed debugging software, a debugging software interface will pop up. Click the "Open Serial Port - Start Sending" buttons in sequence. If the height is displayed in the "Current Height" text box, it indicates successful communication between the development board and the sensor. Clicking the "Start Video Stream" button, if the camera monitoring screen appears, it indicates successful communication between the development board and the camera. Clicking the "Relay Action" button, if the elevator issues an overload alarm (an "Overload" message appears on the display panel along with a buzzer alarm sound), it indicates correct wiring between the development board, relay, and elevator car top control board.

[0055] Sensor module testing: After clicking the "Open Serial Port - Start Sending" buttons in the debugging software interface, the "Current Height" text box will display the current distance value of the sensor, and the "Sensor Connection Status" indicator light will turn green. Then, after clicking the "Set Reference" button, the car is assumed to be unloaded. The "Reference Zero Load Height" text box will display the recorded reference height value, and the "Current Load" text box will display the estimated car load (which should be 0 kg). Adding a 20 kg weight to the car will then cause changes in the readings of the "Current Height" and "Current Load" text boxes. This indicates that the sensor is working normally and its sensitivity meets the requirements for estimating the car load.

[0056] Visual recognition module test: After clicking "Start Video Stream" in the debugging software interface, the camera monitoring screen is displayed and the "Camera Connection Status" indicator light is green. When no one enters the car, the "Is it Empty?" indicator light is green. When someone enters the car, the "Is it Empty?" indicator light is red, and the "Persons" column in the upper left corner of the monitoring screen is not 0. Figure 9 As shown in the image. If the above description is met, it indicates that the visual recognition module is functioning normally.

[0057] An embodiment of the present invention also provides steps for data acquisition and polynomial model construction, as detailed below: Step 1: Car Load Estimation Model (1) Software debugging preparation Open the debugging software interface and click the "Open Serial Port - Start Sending" button in sequence, and wait for the reading in the "Current Height" text box to stabilize.

[0058] (2) Data collection ① Initial reference calibration: Record the initial height value under no-load conditions. .

[0059] ②Step loading: Add 20kg standard weights step by step into the car. After each step, wait for the height to stabilize and record the height corresponding to the load. ,like Figure 6 As shown.

[0060] ③ Data termination condition: When the cumulative load reaches 1100kg (overload threshold 110%), data collection will stop.

[0061] ④ Curve generation: using Plot the load-relative height variation curve with zero point as the reference point. ),like Figure 7 As shown.

[0062] (3) Fitting model A quartic polynomial model is fitted using either the weighted least squares method or Newton's iteration method. The formula for the quartic polynomial model is as follows:

[0063] in, Represents the load inside the car (unit: kg). The data represents the change in height (in mm). The fitting weight for the 800-1000 kg range is increased by 1.5 times to enhance fitting accuracy near the overload zone. Additionally, model accuracy can be improved by collecting multiple sets of data for fitting.

[0064] (4) Initial benchmark calibration and real-time solution ① Baseline Calibration: Take the average unloaded height of each group of data as the unloaded height value H0; take the average full-loaded height (1000kg) of each group of data as the full-loaded height value. (Right now Figure 5 Full_Load in the middle.

[0065] ② Real-time load estimation: This involves estimating the real-time height difference. Input a polynomial model, use 500 as a guess value to iteratively approximate and solve the fourth-order polynomial, and obtain the estimated car load value. .

[0066] The second step is the visual recognition model. (1) Algorithm selection The YOLO (You Only Look Once) object detection algorithm was chosen because it strikes a balance between real-time detection and accuracy, making it particularly suitable for elevator environments that require rapid response. Furthermore, YOLO is easy to optimize and deploy on embedded platforms, which is crucial for resource-constrained devices.

[0067] (2) Dataset creation The dataset collection covered various scenes inside elevators, including different numbers of people, different items, and different lighting conditions. A computer-based annotation platform was built to annotate the images, marking the bounding boxes and categories of the targets. After annotation, the dataset was divided into a training set (70-80%), a validation set (10-15%), and a test set (10-15%) to ensure a balanced distribution of data across all parts.

[0068] (3) Model training The training process utilizes computing servers equipped with high-performance GPUs to handle the computational demands of the deep learning model. Starting with a pre-trained YOLO model, fine-tuning is performed for the elevator scenario. During training, a training set is used, validation set performance is monitored, and hyperparameters (such as learning rate and batch size) are adjusted to prevent overfitting. Finally, the model is evaluated on a test set using metrics such as mean AP, precision, recall, and F1 score to ensure the expected accuracy is achieved.

[0069] (3) Model Deployment The trained YOLO model was converted to a format suitable for embedded platforms. The converted model was tested on a computer to ensure no significant performance degradation. Next, the model was transferred to the embedded platform, the necessary runtime libraries were installed, and it was integrated into the system software. Finally, comprehensive testing was conducted on the embedded platform to verify real-time performance, accuracy, and robustness in various scenarios.

[0070] The following are the steps for implementing the overall system functionality: I. System Startup and Baseline Initialization After completing module deployment, module testing, and data acquisition without any abnormalities, the intelligent weighing program can be started. The average unloaded height of each data set during the data acquisition phase is used as the initial unloaded height value. The average full-load height (1000kg) of each data set is taken as the initial full-load height value. .

[0071] II. Overload Detection and Real-time Monitoring During normal elevator operation, the pull-rope sensor module monitors the real-time height of the car floor rubber. The load inside the car is estimated based on the height difference. Real-time height less than the full load threshold height When the elevator car is determined to be overloaded, the calculation and control module controls the optocoupler isolation relay to activate the elevator's audible and visual alarm via the GPIO pin.

[0072] The elevator car interior is monitored in real time by a network camera, and the number of passengers is counted. When the number of passengers N is greater than the rated number of passengers + 4, or when an electric vehicle enters the car, the car is judged to be in an abnormal situation. The calculation and control module controls the optocoupler isolation relay to activate the elevator's audible and visual alarm through the GPIO pin.

[0073] III. Self-calibration procedure During normal elevator operation, the control calculation module in the development board records the data from the pull-cord sensor when the elevator is in an unloaded stop state and periodically calculates the average zero-load height. If the difference between the average zero-load height and the initial zero-load height (i.e., the offset) is significant, Figure 5 If the equivalent load corresponding to (different_H) is greater than or equal to 50kg or the offset is greater than the preset offset threshold, it is determined that the car floor rubber is aging. The full load height of the car needs to be recalibrated. Then, the old zero load height is updated using the latest average zero load height, and the difference between the two is used to compensate for the full load height. After clearing the recorded data from the pull-cord sensor, recording starts again, and the new zero load height and full load height are used in daily car full load judgment to complete one automatic calibration process.

[0074] Here, for Figure 5 Here are some explanations of the parameters. `avg = Sum / i+1` can be understood as... difference_H = zeroLoad_H - avg, which is the average height under zero load. and zero load height value The offset between the two, abs(EstimateLoad(difference_H,100)), can be understood as the equivalent load corresponding to the difference between the average zero load height and the initial zero load height.

[0075] In summary, this invention employs a lightweight YOLO algorithm deployed on a development board, enabling accurate detection of elevator passengers. It also establishes a mathematical model of the pull-cord sensor and elevator load capacity, and performs weighing calculations based on this model to ensure data accuracy. The pull-cord sensor and development board achieve real-time data interaction via serial communication, combined with an intelligent self-calibration mechanism. When calibration conditions are met, the system automatically triggers the self-calibration mechanism, improving long-term stability. Furthermore, this invention develops a visual interface on the QT platform, supporting local and remote operation. Users can access the QT interface via mobile phone or tablet for remote monitoring and control. For elevator overload alarms, this weight is calculated based on self-calibrated height data. When the full load threshold is exceeded, the development board drives a relay to control the elevator circuit, ensuring safe elevator operation.

[0076] Experimental results show that the proposed method achieves a weighing accuracy of 98.7% and an overload warning response time of less than 50ms, significantly outperforming traditional single-sensor solutions. Furthermore, the multimodal data fusion algorithm ensures efficient computing power, enabling it to maintain high-precision, real-time load monitoring capabilities even during peak hours and in complex load distribution environments. This demonstrates high innovation, practical value, and market potential, and can be widely applied in elevator safety control, providing intelligent safety assurance for elevator manufacturers, property management, and maintenance companies, thereby creating greater economic and social benefits.

[0077] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A multimodal self-calibration method for elevator weighing early warning, characterized in that, The multimodal self-calibration method includes the following steps: A pull-rope sensor is installed between the car frame and the bottom of the car to obtain the zero-load height value when the elevator is in an unloaded and stopped state and the full-load height value when it is in a fully loaded state. Obtain the average zero-load height of multiple zero-load height values ​​within a preset time period, and determine whether the car floor rubber has aged based on the offset between the average zero-load height and the zero-load height values; If the car floor rubber is aged, the old zero-load height value is updated with the latest zero-load height average value, and the full-load height value is compensated and calibrated according to the offset to obtain the latest full-load height value. The multimodal self-calibration method further includes the following steps: The real-time load height value of the elevator is obtained through the pull-rope sensor; The real-time height difference is obtained based on the real-time load height value and the zero load height value; The real-time load of the car is obtained based on the real-time height difference; The specific method for obtaining the real-time load of the car based on the real-time height difference includes the following steps: A polynomial model is constructed based on the car load and the real-time height difference. The polynomial model is fitted, and the real-time load of the car is obtained based on the fitted polynomial model. The polynomial model is expressed as follows: ; in, Indicates the car load. Indicates the real-time height difference. These represent the coefficients of the quartic, cubic, quadratic, and linear terms, as well as a preset constant. According to the formula Compensation calibration is performed on the full load height value; in, Indicates the offset. These represent the zero-load height value and the average zero-load height, respectively. This indicates the full-load height value before compensation calibration. This indicates the full-load height value after compensation calibration. This represents the compensation coefficient.

2. The multimodal self-calibration method for elevator weighing early warning as described in claim 1, characterized in that, The multimodal self-calibration method further includes the following steps: Determine whether the real-time load height value is less than the latest full load height value. If so, determine that the car is overloaded and activate the elevator's audible and visual alarm.

3. The multimodal self-calibration method for elevator weighing early warning as described in claim 1, characterized in that, The main body of the pull-rope sensor is fixedly installed on the central rigid support point at the bottom of the car. The pull-rope axis of the pull-rope sensor is set in the vertical direction and the stretching direction of the pull-rope is consistent with the sinking direction of the car.

4. The multimodal self-calibration method for elevator weighing early warning as described in claim 1, characterized in that, The multimodal self-calibration method further includes the following steps: The system acquires real-time image data of the elevator car's interior, and uses this data to determine the number of passengers and whether any electric vehicles have entered the car. If the number of passengers exceeds a preset threshold or an electric vehicle enters the elevator car, the elevator's audible and visual alarm will be activated.

5. The multimodal self-calibration method for elevator weighing early warning as described in claim 4, characterized in that, Real-time image data is acquired by a network camera installed diagonally on the top of the elevator car away from the car door, and the number of passengers is obtained by identifying passengers based on the YOLO object detection algorithm.

6. The multimodal self-calibration method for elevator weighing early warning as described in claim 5, characterized in that, When the number of passengers detected is zero and the data fed back by the pull-cord sensor is stable and remains at a preset time value, it is determined that the elevator is in an unloaded and stopped state.

Images