Mine hoisting system steel wire rope monitoring method and system based on AI vision

Abstract

The invention discloses a mine hoisting system steel wire rope monitoring method based on AI vision. The method comprises the specific steps that S1, an image sequence is collected; s2, extracting vibration displacement of the steel wire rope; s3, determining a vibration fundamental frequency; s4, calculating the tension of the steel wire rope; and S5, monitoring data output and intelligent diagnosis. According to the method, nonlinear interference and error accumulation of a traditional indirect measurement method are avoided fundamentally; sub-pixel displacement analysis and time sequence filtering are combined, and the dynamic measurement precision can be within + / -2%; single-rope independent monitoring is realized, and the technical blank of the industry is filled; unique intelligent diagnosis capability is provided, and the leap of safety management quality is realized; and full-period dynamic monitoring is realized, and complete data support is provided.

Description

Technical Field

[0001] This invention relates to the field of industrial safety monitoring technology, and in particular to a method and system for monitoring steel wire ropes in mine hoisting systems based on AI vision. Background Technology

[0002] Mine hoisting systems are the main vertical transportation arteries in mine production. As the core load-bearing component, the tension state of the wire rope and the actual load on the hoisting container are the two most critical physical quantities that determine the safety of the system. The current mainstream monitoring methods in the industry, such as the motor current method, the hydraulic sensor method, and the static weighing method, all have insurmountable physical limitations and engineering dilemmas, such as low measurement accuracy, inability to independently monitor the tension of a single rope, and inability to achieve full-cycle dynamic monitoring.

[0003] The fundamental flaws in these methods have led to two long-standing and unresolved core security vulnerabilities becoming particularly prominent:

[0004] Uneven tension in multiple ropes: In a multi-rope friction hoisting system, if the tension difference between the ropes exceeds the safe range, it will drastically accelerate the uneven wear of the wire ropes and friction pads, which is a major and fundamental hidden danger that may induce catastrophic accidents such as rope slippage and rope breakage.

[0005] Skip Residual Ore Overloading: Materials such as sticky ore or frozen coal in winter may not be completely discharged after unloading and remain in the skip, forming residual ore. This results in an additional dangerous and unknown weight on top of the known load during the next loading, constituting a serious overloading.

[0006] Existing patent technologies have also explored this area. For example, CN115018785A discloses a method for detecting the tension of hoisting wire ropes based on visual vibration frequency recognition. However, this method uses traditional Canny edge detection and tilted rectangle fitting techniques. In the harsh environment of mines with high dust, oil pollution, and changing lighting, the stability and accuracy of edge extraction are difficult to guarantee. Its differential threshold dead point cleaning method, which relies on empirical coefficients, is not robust enough. More importantly, this solution stops at the general tension measurement itself and does not involve deep coupling with the hoist control system PLC. Therefore, it cannot achieve intelligent diagnosis of mining-specific risks such as residual ore overload.

[0007] Another patent, CN111044197B, discloses a non-contact cable force testing system based on an unmanned aerial vehicle (UAV) platform. This method is applicable to large civil engineering structures such as bridges. Its core is to calculate cable force by extracting the frequency difference of higher-order modes. However, in scenarios such as mine hoisting systems with low frequency and small amplitude, the higher-order mode signals are weak, difficult to excite, and difficult to accurately pick up. In addition, this solution also does not consider the independent parallel monitoring of multi-rope systems, nor does it interact with PLC operating data in real time, making it impossible to calculate net load or diagnose tension imbalance. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for monitoring wire ropes in mine hoisting systems based on AI vision. This solves the key safety problems that traditional methods cannot effectively diagnose multi-rope tension imbalance and residual ore overload, and realizes the transformation from simple physical quantity measurement to intelligent safety diagnosis.

[0009] The technical solution adopted in this invention is:

[0010] A method for monitoring wire ropes in mine hoisting systems based on AI vision, comprising the following steps:

[0011] S1. Image sequence acquisition:

[0012] Using several industrial cameras, high-speed images are taken of the free vibration sections of each wire rope to obtain a digital image sequence of the wire rope within a preset time window.

[0013] S2. Extraction of wire rope vibration displacement:

[0014] The acquired digital image sequence is processed to accurately extract the displacement time sequence signal that can characterize the lateral vibration characteristics of the wire rope;

[0015] S3. Determination of the fundamental frequency of vibration:

[0016] Frequency domain analysis is performed on displacement time series signals to accurately extract their fundamental vibration frequency;

[0017] S4. Calculation of wire rope tension:

[0018] Based on the physical model of string vibration, the real-time tension of the wire rope is calculated using the determined fundamental frequency of vibration;

[0019] S5. Monitoring data output and intelligent diagnosis:

[0020] The calculated multi-rope tension data is output and stored, and combined with the operating condition parameters obtained in real time from the hoist's main control PLC for advanced intelligent diagnosis.

[0021] Specifically, the industrial camera acquisition parameters in S1 are set as follows: camera frame rate f s Not less than 500 FPS; this selection is based on the Nyquist sampling theorem, and the expected maximum vibration frequency f of the wire rope. max Typically ≤20Hz, camera frame rate f s It must be much greater than twice the maximum vibration frequency; the time window t for each acquisition cycle is set to 2 seconds, thus obtaining N = f s The image sequence of ×t frames I = {I1, I2, ..., I...} NTo eliminate the motion blur caused by complex lighting changes and high-speed movement of steel wire ropes in the mine environment, a near-infrared active stroboscopic flash source with microsecond-level hardware synchronization with camera exposure is used for supplementary lighting.

[0022] Specifically, step S2 is as follows:

[0023] S21. Image preprocessing: For each frame of the sequence, I... k Gaussian filtering is used to suppress random noise, and its kernel function is:

[0024] G(x,y)=[1 / (2πσ 2 )]*exp(-[(x 2 +y 2 ) / (2σ 2 )])

[0025] Where σ is the standard deviation of the Gaussian kernel; then histogram equalization is performed to enhance the contrast between the steel wire rope and the background, and the preprocessed image sequence I′ is output.

[0026] S22, Object Detection and Tracking: A YOLOv5 deep learning object detection model pre-trained on a large number of industrial scene samples is used to process the pre-processed image sequence I′. k The steel wire rope in the image is located pixel-level. The confidence threshold (conf) is set to 0.5, and the intersection-union (IoU) threshold (iou) is set to 0.45. The bounding box B of the steel wire rope is then output. k ={x min ,y min ,x max ,y max The bounding box is used to determine the coordinates z of the center point of the wire rope. k =(x c ,y c This serves as the observation input for subsequent filtering and defines the region of interest (ROI) for subpixel displacement analysis.

[0027] To ensure the continuity and smoothness of inter-frame tracking, a Kalman filter is introduced. The state vector of this filter is defined as X = [p, v]T, where p is the position and v is the velocity. The state transition matrix A and the observation matrix H are respectively:

[0028]

[0029] H = [1, 0]

[0030] Where Δt=1 / f s It is the sampling time interval; through the prediction and update loop of Kalman filtering, a robust and smooth trajectory of the center position p of the wire rope is output. kImage quality is assessed by comparing the confidence level of the tracking results with the confidence threshold conf. Sequences that pass the assessment proceed to the next step of analysis. Matrix A describes the temporal evolution of the system state, i.e., the position at the next moment is equal to the current position plus the product of the velocity and the time interval. Matrix H represents the observation model, i.e., only the position information can be directly observed, and the velocity needs to be estimated through filtering.

[0031] More specifically, the prediction and update loop of the Kalman filter is as follows:

[0032] Prediction: Based on the state X of the previous time step k-1 Predict the current state X using the state transition matrix A. k|k-1 =AX k-1 ;

[0033] Update: The center z-axis of the bounding box obtained by YOLO detection... k The observed values ​​are weighted and fused with the predicted values ​​to obtain the optimal estimate X. k ;

[0034] Output: Extract the position component from the state vector to obtain the smooth trajectory p of the wire rope center position. k ;

[0035] S23. Subpixel displacement analysis: To achieve high measurement accuracy, the phase correlation method is used to calculate the continuous frame I′. k and I′ k+1 The normal small displacement Δd between the centerlines of the wire ropes k ;

[0036] First, the trajectory p of the wire rope center position based on the Kalman filter output. k Define a rectangular region of interest (ROI); the major axis of the ROI is parallel to the direction of the wire rope, and the minor axis is perpendicular to the wire rope, to ensure that the main normal vibration component is captured.

[0037] The phase correlation method is implemented by calculating the inverse Fourier transform of the cross power spectrum of the region of interest (ROI) between two frames of images:

[0038]

[0039] The phase difference between two images in the frequency domain directly corresponds to their spatial displacement. Pure phase information can be extracted by normalizing the cross power spectrum, and the peak position of its inverse Fourier transform (U) can be determined. peak V peak This represents a two-dimensional displacement vector;

[0040] By finding the peak position in the inverse transform result, sub-pixel level displacement is obtained with an accuracy of 0.1 pixels. The displacement values ​​of the entire sequence are concatenated, specifically by taking the minute normal displacement Δd of the wire rope centerline of each frame.k The vibration trajectory is obtained by summing the accumulated data, and then determined based on the observation field length L. obs The camera calibration coefficient K = L is determined by the corresponding number of pixels M. obs The / Mmm / pixel is converted to the physical unit mm, and finally a high-fidelity displacement time series signal d(t) is generated.

[0041] Since the minor axis of the ROI is set perpendicular to the wire rope, the v component at the peak position corresponds to the normal minute displacement Δd. k The u-component parallel to the rope direction has a very small value during normal vibration and is not considered a monitoring target. Thus, by selecting a directional area of ​​interest (ROI) and extracting its components, accurate measurement of normal vibration is achieved.

[0042] Specifically, step S3 is as follows:

[0043] S31: Signal window function processing: To reduce spectral leakage, the displacement time series signal d(t) is first processed using a Hanning window;

[0044] S32: Frequency domain transformation: Apply Fast Fourier Transform (FFT) to convert the time domain signal into a frequency domain signal to obtain the vibration spectrum Y(f);

[0045] S33: Fundamental Frequency Extraction and Verification: Within the estimated frequency range, a robust peak detection algorithm is used to search for the most significant energy peak. To ensure that the extracted frequency is the fundamental frequency f1 and not its harmonics, the algorithm includes harmonic verification logic. The fundamental frequency value is considered valid only when the signal quality meets the requirements, i.e., checking for the existence of a significant energy f1 / 2 component and confirming harmonic relationships such as f2≈2f1, thereby accurately locking the fundamental frequency f1. Simultaneously, a threshold is set to calculate the signal-to-noise ratio of the spectrum.

[0046] SNR. This step adds harmonic verification and SNR quality control, making it more reliable in high-noise mine environments compared to existing technologies.

[0047] Specifically, step S4 is as follows:

[0048] The physical model of string vibration is derived from the wave equation of string vibration:

[0049]

[0050] Under the boundary conditions of fixed ends, its fundamental frequency f1 is related to the tension t, the mass per unit length μ, and the effective chord length L. eff The relationship is:

[0051]

[0052] By rearranging terms, the formula for calculating tension can be obtained:

[0053]

[0054] Where μ is a known parameter of the wire rope, in units of kg / m and L. eff The effective length of the observation segment calibrated by the camera is expressed in meters; this process is performed independently and in parallel for each wire rope, thereby achieving independent monitoring of the tension of a single rope.

[0055] Specifically, step S5 is as follows:

[0056] Precise calculation of net load: When the PLC obtains information that the hoisting container is operating at a constant speed, the system calculates the total tension ∑T of all wire ropes. i Subtract the known weight W of the lifting container converyance and the weight of the suspended wire rope (μgH) calculated based on the real-time depth. current This allows for the accurate calculation of the net load W inside the container. payload :

[0057]

[0058] Where n is the number of steel wire ropes, and g is the acceleration due to gravity;

[0059] Intelligent diagnosis of residual ore overload: When the operating status word S obtained from the PLC indicates that the hoisting container is in a specific working condition of being lowered under no-load after unloading, the system automatically performs the above-mentioned net load calculation; if the calculated W payload If the residual ore safety threshold is exceeded, it is determined that residual ore overload has occurred and a high-level alarm is immediately triggered.

[0060] Intelligent tension imbalance diagnosis: The system calculates the independent tension T of each wire rope in real time through parallel processing. i And continuously calculate the imbalance I mb Imbalance can be defined as:

[0061]

[0062] Among them, T max T min and T avg These represent the maximum, minimum, and average tension values ​​for all wire ropes; when the unbalance is I... mb If the tension exceeds the threshold defined by the safety regulations for an extended period, it is determined that uneven tension has occurred, and a corresponding warning or alarm is triggered.

[0063] A mine hoisting system wire rope monitoring system based on AI vision, comprising:

[0064] Image acquisition module; including at least one high-speed industrial camera and an active stroboscopic light source that is synchronized with it at the microsecond level;

[0065] Central processing module; an industrial computer with an embedded CPU and high-performance GPU, used to efficiently execute all vision algorithms and solve physical models;

[0066] Data interface module; used for stable and reliable two-way communication with the PLC of the mine hoist;

[0067] Storage and execution module; containing a non-transitory computer-readable medium storing executable instructions.

[0068] Due to the adoption of the technical solution described above, the present invention has the following advantages:

[0069] This invention directly measures the deterministic physical relationship between tension and T = 4 μL. 2 f1 2 The fundamental frequency of the steel wire rope vibration is determined, which fundamentally avoids the nonlinear interference and error accumulation of traditional indirect measurement methods; combined with subpixel displacement analysis and time-series filtering, the dynamic measurement accuracy can reach within ±2%.

[0070] Achieving independent monitoring of a single rope fills a technological gap in the industry: By configuring an independent visual perception and analysis path for each wire rope, independent, parallel, and real-time monitoring of the tension of each rope in a multi-rope system has been achieved for the first time in engineering, technically solving the long-standing fundamental problem of diagnosing uneven tension in multiple ropes.

[0071] It possesses unique intelligent diagnostic capabilities, achieving a qualitative leap in safety management: its original intelligent diagnostic algorithm for residual ore overload and uneven tension, which is deeply coupled with PLC operating conditions, can accurately identify and warn of fatal safety hazards that cannot be detected by two traditional methods, elevating safety management from post-event analysis to a new level of pre-event warning and process control.

[0072] Non-contact, non-intrusive deployment with excellent engineering adaptability and safety: The system installation does not require any invasive modifications to the existing mechanical and electrical main systems of the hoist, does not change the stress path of the original system, does not introduce new failure points, and has low implementation risk, controllable cost, and short cycle.

[0073] Achieve full-cycle dynamic monitoring and provide complete data support: It can continuously collect and analyze data on the complete operating cycle of the hoist from acceleration, constant speed to deceleration, providing an unprecedented high-value full-cycle load map for equipment health management, fault tracing and digital twin construction. Attached Figure Description

[0074] Figure 1 This is a schematic diagram of the overall deployment of the monitoring system of the present invention in a mine shaft environment;

[0075] In the diagram: 100-well tower, 110-wire rope, 112-friction wheel, 114-head sheave, 116-lifting container, 120-high-speed camera, 130-strobe light source, 200-industrial computer (IPC).

[0076] Figure 2 This is a hardware architecture block diagram of the present invention.

[0077] Figure 3 This is a schematic diagram of the workflow of the present invention.

[0078] Figure 4 This is a conceptual schematic diagram of the steel wire rope from image analysis to spectrum extraction according to the present invention.

[0079] Figure 5 This is an example of the main monitoring dashboard of the human-machine interface (HMI) of the present invention.

[0080] Figure 6 This is a logic flowchart of the intelligent diagnosis of the present invention. Detailed Implementation

[0081] The present invention will be further explained and described below with reference to the accompanying drawings and embodiments. However, this should not be construed as limiting the scope of protection of the present invention. The purpose of disclosing the present invention is to protect all technical improvements within the scope of the present invention.

[0082] A method for monitoring wire ropes in mine hoisting systems based on AI vision, comprising the following steps:

[0083] S1. Image sequence acquisition:

[0084] Several industrial cameras were used to capture high-speed images of the free vibration sections of each wire rope, obtaining a digital image sequence of the wire rope within a preset time window. The industrial camera acquisition parameters were set as follows: camera frame rate f. s Not less than 500 FPS; this selection is based on the Nyquist sampling theorem, and the expected maximum vibration frequency f of the wire rope. max Typically ≤20Hz, camera frame rate f s It must be much greater than twice the maximum vibration frequency; the time window t for each acquisition cycle is set to 2 seconds, thus obtaining N = f s The image sequence of ×t frames I = {I1, I2, ..., I...} N To eliminate the motion blur caused by complex lighting changes and high-speed movement of steel wire ropes in the mine environment, a near-infrared active stroboscopic flash source with microsecond-level hardware synchronization with camera exposure is used for supplementary lighting.

[0085] S2. Extraction of wire rope vibration displacement:

[0086] The acquired digital image sequence is processed to accurately extract the displacement time series signal that characterizes the lateral vibration characteristics of the wire rope; specifically:

[0087] S21. Image preprocessing: For each frame of the sequence, I... k Gaussian filtering is used to suppress random noise, and its kernel function is:

[0088] G(x,y)=[1 / (2πσ 2 )]*exp(-[(x 2 +y 2 ) / (2σ 2 )])

[0089] Where σ is the standard deviation of the Gaussian kernel; then histogram equalization is performed to enhance the contrast between the steel wire rope and the background, and the preprocessed image sequence I′ is output.

[0090] S22, Object Detection and Tracking: A YOLOv5 deep learning object detection model pre-trained on a large number of industrial scene samples is used to process the pre-processed image sequence I′. k The steel wire rope in the image is located pixel-level. The confidence threshold (conf) is set to 0.5, and the intersection-union (IoU) threshold (iou) is set to 0.45. The bounding box B of the steel wire rope is then output. k ={x min ,y min ,x max ,y max The bounding box is used to determine the coordinates z of the center point of the wire rope. k =(x c ,y c This serves as the observation input for subsequent filtering and defines the region of interest (ROI) for subpixel displacement analysis.

[0091] To ensure the continuity and smoothness of inter-frame tracking, a Kalman filter is introduced. The state vector of this filter is defined as X = [p, v]T, where p is the position and v is the velocity. The state transition matrix A and the observation matrix H are respectively:

[0092]

[0093] H = [1, 0]

[0094] Where Δt=1 / f s It is the sampling time interval; through the prediction and update loop of Kalman filtering, a robust and smooth trajectory of the center position p of the wire rope is output. kImage quality is assessed by comparing the confidence level of the tracking results with the confidence threshold conf. Sequences that pass the assessment proceed to the next step of analysis. Matrix A describes the temporal evolution of the system state, i.e., the position at the next moment is equal to the current position plus the product of the velocity and the time interval. Matrix H represents the observation model, i.e., only the position information can be directly observed, and the velocity needs to be estimated through filtering.

[0095] More specifically, the prediction and update loop of the Kalman filter is as follows:

[0096] Prediction: Based on the state X of the previous time step k-1 Predict the current state X using the state transition matrix A. k|k-1 =AX k-1 ;

[0097] Update: The center z-axis of the bounding box obtained by YOLO detection... k The observed values ​​are weighted and fused with the predicted values ​​to obtain the optimal estimate X. k ;

[0098] Output: Extract the position component from the state vector to obtain the smooth trajectory p of the wire rope center position. k .

[0099] S23. Subpixel displacement analysis: To achieve high measurement accuracy, the phase correlation method is used to calculate the continuous frame I′. k and I′ k+1 The normal small displacement Δd between the centerlines of the wire ropes k ;

[0100] First, the trajectory p of the wire rope center position based on the Kalman filter output. k Define a rectangular region of interest (ROI); the major axis of the ROI is parallel to the direction of the wire rope, and the minor axis is perpendicular to the wire rope, to ensure that the main normal vibration component is captured.

[0101] The phase correlation method is implemented by calculating the inverse Fourier transform of the cross power spectrum of the region of interest (ROI) between two frames of images:

[0102]

[0103] The phase difference between two images in the frequency domain directly corresponds to their spatial displacement. Pure phase information can be extracted by normalizing the cross power spectrum, and the peak position of its inverse Fourier transform (U) can be determined. peak V peak This represents a two-dimensional displacement vector;

[0104] By finding the peak position in the inverse transform result, sub-pixel level displacement is obtained with an accuracy of 0.1 pixels. The displacement values ​​of the entire sequence are concatenated, specifically by taking the minute normal displacement Δd of the wire rope centerline of each frame.k The vibration trajectory is obtained by summing the accumulated data, and then determined based on the observation field length L. obs The camera calibration coefficient K = L is determined by the corresponding number of pixels M. obs The / Mmm / pixel is converted to the physical unit mm, and finally a high-fidelity displacement time series signal d(t) is generated.

[0105] Since the minor axis of the ROI is set perpendicular to the wire rope, the v component at the peak position corresponds to the normal minute displacement Δd. k The u-component parallel to the rope direction has a very small value during normal vibration and is not considered a monitoring target. Thus, by selecting a directional area of ​​interest (ROI) and extracting its components, accurate measurement of normal vibration is achieved.

[0106] S3. Determination of the fundamental frequency of vibration:

[0107] Frequency domain analysis is performed on the displacement time series signal to accurately extract its fundamental vibration frequency; specifically:

[0108] S31: Signal window function processing: To reduce spectral leakage, the displacement time series signal d(t) is first processed using a Hanning window;

[0109] S32: Frequency domain transformation: Apply Fast Fourier Transform (FFT) to convert the time domain signal into a frequency domain signal to obtain the vibration spectrum Y(f);

[0110] S33: Fundamental Frequency Extraction and Verification: Within the estimated frequency range, a robust peak detection algorithm is used to search for the most significant energy peak. To ensure that the extracted frequency is the fundamental frequency f1 and not its harmonics, the algorithm includes harmonic verification logic. The fundamental frequency value is considered valid only when the signal quality meets the requirements, i.e., checking for the existence of a significant energy f1 / 2 component and confirming harmonic relationships such as f2≈2f1, thereby accurately locking the fundamental frequency f1. Simultaneously, a threshold is set to calculate the signal-to-noise ratio of the spectrum.

[0111] SNR. This step adds harmonic verification and SNR quality control, making it more reliable in high-noise mine environments compared to existing technologies.

[0112] S4. Calculation of wire rope tension:

[0113] Based on the physical model of string vibration, the real-time tension of the wire rope is calculated using the determined fundamental frequency of vibration; specifically:

[0114] The physical model of string vibration is derived from the wave equation of string vibration:

[0115]

[0116] Under the boundary conditions of fixed ends, its fundamental frequency f1 is related to the tension T, the mass per unit length μ, and the effective chord length L. effThe relationship is:

[0117]

[0118] By rearranging terms, the formula for calculating tension can be obtained:

[0119]

[0120] Where μ is a known parameter of the wire rope, in units of kg / m and L. eff The effective length of the observation segment calibrated by the camera is expressed in meters; this process is performed independently and in parallel for each wire rope, thereby achieving independent monitoring of the tension of a single rope.

[0121] S5. Monitoring data output and intelligent diagnosis:

[0122] The calculated multi-rope tension data is output and stored, and combined with the operating condition parameters obtained in real time from the hoist's main control PLC for advanced intelligent diagnostics; specifically:

[0123] Precise calculation of net load: When the PLC obtains information that the hoisting container is operating at a constant speed, the system calculates the total tension ∑T of all wire ropes. i Subtract the known weight W of the lifting container converyance and the weight of the suspended wire rope (μgH) calculated based on the real-time depth. current This allows for the accurate calculation of the net load W inside the container. payload :

[0124]

[0125] Where n is the number of steel wire ropes and g is the acceleration due to gravity.

[0126] Intelligent diagnosis of residual ore overload: When the operating status word S obtained from the PLC indicates that the hoisting container is in a specific working condition of being lowered under no-load after unloading, the system automatically performs the above-mentioned net load calculation; if the calculated W payload If the residual ore safety threshold is exceeded, it is determined that residual ore overload has occurred, and a high-level alarm is immediately triggered.

[0127] Intelligent tension imbalance diagnosis: The system calculates the independent tension T of each wire rope in real time through parallel processing. i And continuously calculate the imbalance I mb Imbalance can be defined as:

[0128]

[0129] Among them, T max T min and T avgThese represent the maximum, minimum, and average tension values ​​for all wire ropes; when the unbalance is I... mb If the tension exceeds the threshold defined by the safety regulations for an extended period, it is determined that uneven tension has occurred, and a corresponding warning or alarm is triggered.

[0130] The following is in conjunction with the appendix Figure 3 and 4 The monitoring method and algorithm flow of the present invention are described below:

[0131] Figure 3 The core monitoring method flow running on the IPC (200) is shown in detail. This flow begins with image sequence acquisition (310); the IPC (200) triggers all cameras (120) to synchronously acquire a 2-second image sequence at a frame rate of 500 FPS via the synchronization unit (250); after acquisition, the displacement extraction process begins; such as... Figure 4 As shown, the acquired image sequence (410) is first preprocessed (320); then, wire rope detection and tracking are performed using YOLOv5 and Kalman filtering (330), and image quality is evaluated (335); for sequences with qualified quality, subpixel displacement analysis is performed using phase correlation method (340), and finally a high-fidelity displacement time sequence signal (420) is generated.

[0132] After obtaining the displacement time series signal (420), it is subjected to FFT frequency domain analysis (350) to obtain the vibration spectrum (430); then, the system performs signal validity test (355), and accurately identifies the effective vibration fundamental frequency f1 (440) through peak detection, harmonic verification and SNR quality assessment.

[0133] After identifying the fundamental frequency, perform tension and load calculation (360); substitute the effective fundamental frequency f1 into the physical formula T=4μL2f12 to accurately calculate the real-time tension T of each wire rope; finally, perform monitoring data output and storage (370).

[0134] Combined with appendix Figure 5 and 6 The intelligent diagnosis and application of the present invention will be explained as follows: the calculated independent tension data is the core foundation for realizing the unique diagnostic function of the present invention; refer to the appendix. Figure 5 The HMI (500) visually displays the monitoring results in the form of a multi-rope tension radar map (520) and a load-depth real-time curve (510);

[0135] Reference Figure 6 The diagnostic logic flow involves the diagnostic module starting in parallel after acquiring real-time data (610):

[0136] Tension unevenness diagnosis: The system continuously judges tension unevenness (620). If the imbalance exceeds the threshold of 10%, a tension unevenness warning is triggered (630).

[0137] Residual ore overload diagnosis: The system determines whether it is in the unloaded and unloaded condition based on the PLC status word (640); if so, it calculates the net load at this moment. If the value exceeds the residual ore threshold of 500kg, residual ore overload judgment is performed (650), and a serious alarm for residual ore overload is triggered (660).

[0138] All diagnostic results and raw data are recorded and stored (670) for subsequent trend analysis (680) and automated report generation (690). By implementing this invention, the safety management of mine hoisting systems can be transformed into a proactive, data-driven, and predictable new model.

[0139] The present invention discloses an AI vision-based wire rope monitoring system for mine hoisting systems. The monitoring system is deployed inside the mine shaft tower (100) and includes:

[0140] Image acquisition module; including at least one high-speed industrial camera (120) and an active stroboscopic light source (130) synchronized with it at the microsecond level;

[0141] Central processing module; an industrial computer IPC (200) with an embedded CPU and high-performance GPU, used to efficiently execute all vision algorithms and physical model solutions;

[0142] Data interface module (260); used for stable and reliable bidirectional communication with the PLC (270) of the mine hoist;

[0143] Storage and execution module; containing a non-transitory computer-readable medium storing executable instructions.

[0144] For each steel wire rope (110) to be monitored, an independent image acquisition module is deployed, including a high-speed industrial camera (120) and an active stroboscopic light source (130); the field of view of the high-speed industrial camera (120) covers a free rope segment with a length L of 1.5-3 meters located between the friction wheel (112) and the sheave (114); the reasonable installation height H (15-20 meters) and observation distance D (2-4 meters) ensure high-quality imaging.

[0145] Reference Figure 2All high-speed industrial cameras (120) are connected to a high-performance industrial computer (IPC) (200) serving as the central processing module via industrial Ethernet. The hardware configuration of the IPC (200) includes a multi-core CPU (210), such as an Intel Xeon E-2278G, a GPU (220) for parallel computing acceleration, such as an NVIDIA RTX A4000, large-capacity ECC memory (230), and a high-speed solid-state drive (240). The IPC (200) communicates with the main control PLC (270) of the hoist via a data interface module (260) using the OPC UA protocol to obtain real-time operating information such as the depth and speed of the hoisting container (116). The entire system is protected by an uninterruptible power supply (UPS) to ensure its stable and reliable operation.

[0146] The parts of this invention not described in detail are prior art.

[0147] The embodiments selected herein for the purpose of disclosing the inventive objectives are currently considered suitable; however, it should be understood that the invention is intended to include all variations and modifications of the embodiments that fall within the scope of this concept and invention.

Claims

1. An AI vision-based monitoring method for a mine hoist system steel wire rope, characterized in that: The specific steps are: S1, image sequence acquisition: Through several industrial cameras, the free vibration section of each steel wire rope is high-speed photographed, and the digital image sequence of the steel wire rope in the preset time window is obtained; S2, steel wire rope vibration displacement extraction: The collected digital image sequence is calculated to accurately extract the displacement time sequence signal representing the transverse vibration characteristics of the steel wire rope; S3, vibration fundamental frequency determination: The displacement time sequence signal is subjected to frequency domain analysis to accurately extract the vibration fundamental frequency; S4, steel wire rope tension calculation: According to the string vibration physical model, the real-time tension of the steel wire rope is calculated by using the determined vibration fundamental frequency; S5, monitoring data output and intelligent diagnosis: The calculated multi-rope tension data is output and stored, and combined with the running condition parameters obtained from the main control PLC of the hoist in real time, advanced intelligent diagnosis is carried out.

2. The AI vision-based monitoring method of a mine hoist system steel wire rope according to claim 1, characterized in that, The S1 industrial camera acquisition parameter setting is as follows: camera frame rate f s not less than 500 FPS; the time window t of each acquisition cycle is set to 2 seconds, so as to obtain N = f s × t frames of image sequence I = {I1, I2,..., I N}; and a near-infrared active frequency flash light source is used for light supplement. 3.The AI vision-based monitoring method for a mine hoist steel rope system according to claim 2, characterized in that: The step S2 is specifically: S21, image pre-processing: for each frame image I in the sequence k Gaussian filter is applied to suppress random noise, whose kernel function is: ； wherein σ is the standard deviation of the Gaussian kernel; and then histogram equalization is performed to enhance the contrast between the steel wire rope and the background, outputting a pre-processed image sequence I′ ; S22, target detection and tracking: a YOLOv5 deep learning target detection model pre-trained on a large number of industrial scene samples is used to perform pixel-level positioning on the steel wire in the pre-processed image I' k , with a confidence threshold conf set to 0.5 and an intersection over union threshold iou set to 0.45, to output the bounding box of the steel wire ; the bounding box is used to determine the center point coordinates of the steel wire , as the observation input for subsequent filtering processing, and to define the region of interest ROI for sub-pixel displacement analysis; To ensure the continuity and smoothness of inter-frame tracking, a Kalman filter is introduced, the state vector of which is defined as wherein is the position, is the velocity; the state transition matrix A and the observation matrix H are respectively ; wherein, is a sampling time interval; prediction and update loops by Kalman filtering a loop, outputting a robust and smooth wire rope center position trajectory p k ; According to the confidence of the tracking result and the confidence threshold value conf, the image quality is evaluated, and the qualified sequence enters the next step of analysis; S23. Subpixel displacement analysis: To achieve high measurement accuracy, the phase correlation method is used to calculate the displacement of consecutive frames. and The normal small displacement of the centerline of the wire rope between ; First, based on the Kalman filter output of the steel wire rope center position trajectory p k , a rectangular region of interest ROI is defined; the long axis of the ROI is parallel to the steel wire rope direction, and the short axis is perpendicular to the steel wire rope, ensuring that the main normal vibration component is captured; The phase correlation method realizes by calculating the inverse Fourier transform of the cross power spectrum of the two image regions of interest ROI: ； The phase difference between two images in the frequency domain directly corresponds to their spatial displacement. Pure phase information can be extracted by normalizing the cross power spectrum, and the peak position of its inverse Fourier transform ( , This represents a two-dimensional displacement vector; By finding the peak position in the inverse transform result, the sub-pixel level displacement is obtained, and the accuracy can reach 0.1 pixels; the displacement values of the whole sequence are connected in series, specifically the normal small displacement of the center line of the steel wire of each frame The cumulative sum is obtained to obtain the vibration trajectory, and the length of the observation field of view is determined according to the length of the observation field of view The number of pixels corresponding to it The determined camera calibration coefficient Convert to physical units mm, finally generate high-fidelity displacement time sequence signal .

4. The AI vision-based monitoring method of a mine hoist system steel wire rope according to claim 3, characterized in that: The prediction and update cycle of the Kalman filter is specifically: Prediction: based on the state X of the previous time instant k−1 , the state of the current time instant is predicted by the state transition matrix A ; Update: the center of the bounding box obtained by YOLO detection As an observation value, the predicted value is weighted and fused to obtain the optimal estimate X k ; Output: Extract the position component in the state vector to get the smoothed wire center position trajectory p k .

5. The AI vision-based monitoring method of a mine hoist system steel wire rope according to claim 4, characterized in that: The step S3 is specifically: S31: signal window function processing: in order to reduce the spectral leakage, first, the displacement time series signal is processed by using the Hanning window; S32: frequency domain transform: apply fast Fourier transform (FFT) to convert the time domain signal into a frequency domain signal to obtain a vibration frequency spectrum ; S33: Fundamental extraction and verification: In the estimated frequency range, search the most significant peak by robust peak detection algorithm; to ensure that the extracted is the fundamental frequency rather than its harmonics, the algorithm contains harmonic verification logic, only when the signal quality meets the requirements, the fundamental frequency value is considered valid, that is, to test whether there is a significant energy component, and confirm the harmonic relationship, so as to accurately lock the fundamental frequency ; At the same time, set the threshold to calculate the signal-to-noise ratio SNR of the spectrum.

6. The AI vision-based monitoring method of a mine hoist system steel wire rope according to claim 5, characterized in that: The step S4 is specifically: The string vibration physical model is derived from the string vibration wave equation: ； Under the boundary condition of both ends fixed, its fundamental frequency is related to tension , mass per unit length and effective string length as ； Through moving term transformation, the tension calculation formula is obtained: ； wherein, is a known parameter of the steel wire rope, in kg / m, is the effective length of the observation section by camera calibration, in m; this process is performed independently and in parallel for each steel wire rope, thus enabling independent monitoring of the tension of the individual ropes.

7. The AI vision-based monitoring method of a mine hoist system steel wire rope according to claim 6, characterized in that: The step S5 is specifically: Net load accurate calculation: when the PLC obtains that the lifting container is in uniform speed running condition, the system synthesizes the total tension of all steel wires , subtracts the known self weight of the lifting container and the calculated weight of the suspended steel wire according to the real-time depth , thereby accurately calculating the net load in the container : ； wherein, N is the number of steel wire ropes, g is the acceleration of gravity; Intelligent diagnosis of residual ore overload: When the operating status word obtained from the PLC This indicates that when the lifting container is in a specific working condition of being lowered under no-load after unloading, the system automatically performs the above net load calculation; if the calculated net load is... If the residual ore safety threshold is exceeded, it is determined that residual ore overload has occurred and a high-level alarm is immediately triggered. Tension imbalance intelligent diagnosis: the system calculates the independent tension of each steel wire in real time through parallel processing channels and continuously calculates the imbalance degree ; the imbalance degree can be defined as: ； wherein, , and are the maximum, minimum and average values of all steel wire rope tensions, respectively; when the unbalance continues beyond a threshold value defined by the safety regulations, then a tension imbalance is determined to have occurred and the corresponding pre-warning or alarm is triggered.

8. An AI vision-based mine hoist system steel wire rope monitoring system, characterized in that, It includes: Image acquisition module; Including at least one high-speed industrial camera and one active frequency flash light source with microsecond level synchronization; Central processing module; It is an industrial computer with embedded CPU and high-performance GPU, which is used for efficient execution of all visual algorithms and physical model calculation; Data interface module; It is used for stable and reliable two-way communication with the PLC of the mine hoist; Storage and execution module; It contains a non-transient computer readable medium with executable instructions.

Images