Machine learning based safety line fall risk prediction and automatic shutdown method

Abstract

The application belongs to the technical field of safety control of working equipment, and discloses a safety rope falling risk prediction and early warning and automatic shutdown method based on machine learning. In a hanging basket, a working platform and other working equipment that rely on a safety rope for bearing, traction or limiting, the running state of the safety rope directly affects the stability of the bearing body and the safety of work. When the equipment is running, the force, length change and motion state of the safety rope are usually related to the motion state of the bearing body and the action state of the actuator. The existing technology usually adopts mechanical limiting, braking locking, manual inspection or threshold alarm based on a single monitoring quantity, and usually takes the safety rope tension, displacement, speed or a certain actuator state as an independent basis for judgment, and then alarms, slows down or stops the control after detecting the abnormality. The above-mentioned method is not good at utilizing the continuous correlation between the safety rope state, the motion state of the bearing body and the response state of the actuator, and it is difficult to stably judge the gradual evolution process of the abnormal risk.

Description

Technical Field

[0001] This invention belongs to the field of safety control technology for work equipment, specifically a method for predicting, warning, and automatically shutting down safety rope fall risks based on machine learning. Background Technology

[0002] In suspended platforms, work platforms, and other work equipment that rely on safety ropes for load-bearing, traction, or limiting, the operating state of the safety rope directly affects the operational stability and safety of the load-bearing structure. During equipment operation, there is usually a linkage between the tension, length changes, and movement speed of the safety rope and the displacement, speed, and posture of the load-bearing structure. The operating state of the drive mechanism, braking mechanism, and locking mechanism also directly affects the release process of the safety rope. When the safety rope is subjected to abnormal force, the movement of the load-bearing structure is mismatched, or the actuator responds abnormally, the load-bearing structure may become unstable, slide abnormally, or stop incompletely.

[0003] In existing technologies, safety control of such equipment often adopts mechanical limit switches, braking locks, manual inspections, or threshold alarms based on a single monitoring quantity. These methods typically use the tension of the safety rope, displacement, speed, or the state of a certain actuator as independent judgment criteria, and then trigger alarms, deceleration, or shutdown control after an anomaly is detected. Although some control methods can implement braking or locking after an anomaly occurs, they mainly respond to obvious anomalies that have already occurred, and do not make sufficient use of the continuous state changes during equipment operation.

[0004] In practical applications, the main problems with the above methods are: there is a lack of unified analysis on the continuous correlation between the state of the safety rope, the motion state of the carrier, and the response state of the actuator, making it difficult to make stable judgments on the gradual evolution of abnormal risks; at the same time, after the shutdown control is issued, only the shutdown action is triggered, while the confirmation of whether the drive output has sufficiently decayed, whether the safety rope has truly entered a stable state, and whether the actuator has indeed been in position is still insufficient, thus there is a risk that abnormal motion may remain after shutdown. Summary of the Invention

[0005] The purpose of this invention is to provide a method for predicting, warning, and automatically stopping the fall risk of safety ropes based on machine learning, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a safety rope fall risk prediction, early warning, and automatic shutdown method based on machine learning, applicable to work equipment configured with a safety rope, a drive mechanism, a controller, and one of a braking mechanism and a locking mechanism, or simultaneously configured with a braking mechanism and a locking mechanism; during equipment operation, operating data is collected according to a set sampling period, the operating data including safety rope status data, carrier motion data, and actuator status data corresponding to the actual configuration of the equipment.

[0007] After preprocessing the collected operational data, the data is organized according to a sliding time window of a set length to form a continuous time series sample consisting of the current time window and one or more historical time windows preceding the current time window. Based on the continuous time series sample, fall risk characterization information is generated. The fall risk characterization information includes load change characteristics reflecting changes in the load on the safety rope, motion matching characteristics reflecting the correspondence between the movement of the load-bearing body and the safety rope, and response consistency characteristics reflecting the response of the actuator.

[0008] Subsequently, the fall risk characterization information is input into a pre-trained neural network model to obtain the fall risk value corresponding to the current time window. The controller determines the current control mode based on the risk interval where the fall risk value is located and the number of consecutive occurrences of the fall risk value in multiple consecutive time windows, and outputs a control signal corresponding to the control mode.

[0009] When the current control mode is the shutdown mode, the controller first controls the drive mechanism to stop output; if the equipment is equipped with a braking mechanism, it controls the braking mechanism to enter the braking state; if the equipment is equipped with a locking mechanism, it controls the locking mechanism to enter the locking state; if the equipment is equipped with both a braking mechanism and a locking mechanism, it controls the braking mechanism and the locking mechanism to act synchronously to limit the safety rope from being released further.

[0010] After entering the shutdown mode, the system continues to collect shutdown response data within the set response time. Combined with changes in the output of the drive mechanism, changes in the movement of the safety rope, and changes in the status of the actuators corresponding to the actual configuration, the system determines whether the equipment meets the set shutdown safety conditions.

[0011] If the set shutdown safety conditions are not met, the aforementioned control state of continuing to release the safety rope is maintained, and the equipment is prohibited from re-entering the operating state. Thus, risk assessment, shutdown execution, and post-shutdown safety confirmation are incorporated into the same control process, enabling the controller to continue to verify whether the equipment has entered a controlled state after issuing a shutdown control, thereby avoiding incomplete shutdown caused by ending control based solely on a single risk assessment.

[0012] In the technical solution of this invention, the carrier is a basket, work platform, or load component connected to and moving with the safety rope. To ensure that the fall risk assessment is based on data directly corresponding to the actual operating state of the equipment, the collected operating data includes safety rope status data, carrier motion data, and actuator status data. Specifically, the safety rope status data includes safety rope tension data, safety rope length change data, and safety rope speed data; the carrier motion data includes carrier displacement data, carrier velocity data, carrier acceleration data, and carrier posture data; and the actuator status data includes drive output status data, as well as braking mechanism status data, locking mechanism status data, control command status data, and execution feedback status data corresponding to the actual equipment configuration. When processing the operating data, data from different sampling frequencies are first converted to data with a unified sampling frequency. Then, abnormal pulse data is removed or smoothed, missing sampling points are filled in, and data from different physical quantities are converted into standardized data for model calculation. After the above processing, data from different sources can participate in subsequent calculations under a unified time reference, enabling the formed continuous time-series samples to reflect the state changes of the equipment during continuous operation.

[0013] After the continuous time-series samples are formed, fall risk characterization information is generated based on the data within the same continuous time-series sample. The fall risk characterization information includes load change characteristics, motion matching characteristics, and response consistency characteristics. Among them, the load change characteristics are generated from the safety rope tension data, specifically the safety rope tension fluctuation amplitude, safety rope tension change rate, and safety rope tension mutation amount. The motion matching characteristics are generated based on the corresponding sampling point differences and changes of the time-synchronized safety rope length change data, safety rope movement speed data, load displacement data, load velocity data, and load acceleration data, specifically the deviation values ​​between the safety rope length change and the load displacement change, the deviation values ​​between the safety rope movement speed and the load velocity, and the deviation values ​​between the safety rope movement speed change and the load acceleration. The response consistency characteristics are generated based on the corresponding sampling point differences and changes of the time-synchronized control command status data, drive output status data, and braking mechanism status data, locking mechanism status data, and execution feedback status data corresponding to the actual equipment configuration, specifically the drive output change deviation value, braking action feedback deviation value, and locking action feedback deviation value after the control command is issued. By incorporating the changes in the force on the safety rope, the motion relationship of the load-bearing body, and the response of the actuator into the fall risk characterization information, the input of the neural network model can cover the main influencing factors in the risk formation process, thereby avoiding judgment based solely on a single instantaneous signal.

[0014] To enable the neural network model to distinguish between different risk states, the training phase employs normal operation samples, abnormal evolution samples, and abnormal shutdown response samples. Normal operation samples correspond to operating processes where the safety rope state, carrier motion state, and actuator state are all within their respective preset threshold ranges. Abnormal evolution samples correspond to operating processes where at least one of the following exceeds its respective preset threshold range: safety rope force change, deviation between safety rope motion and carrier motion, or actuator response lag. Abnormal shutdown response samples correspond to operating processes where the set shutdown safety conditions are not met even after the equipment executes shutdown control.

[0015] When training the neural network model using the above three types of samples, the fall risk representation information corresponding to the continuous time series samples is used as the model input, and the labeled results corresponding one-to-one with the normal operation samples, abnormal evolution samples and shutdown response abnormal samples are respectively assigned as the model output targets. By adopting this sample organization and labeling method, the neural network model can establish a correspondence between continuous operation state and risk level, thereby improving the ability to identify abnormal evolution process and shutdown failure state.

[0016] During the control phase, the risk intervals are set as the first, second, and third intervals, with the risk level increasing sequentially. The control mode is progressively judged based on the number of consecutive occurrences of the fall risk value within multiple consecutive time windows. First, it is determined whether the corresponding conditions for the shutdown mode are met. If the corresponding conditions for the shutdown mode are not met, it is determined whether the corresponding conditions for the speed limit mode are met. If the corresponding conditions for both the shutdown mode and the speed limit mode are not met, it is determined whether the corresponding conditions for the warning mode are met. If none of the above conditions are met, the operation mode is determined.

[0017] By employing the aforementioned progressive judgment method, we can prevent higher risk values ​​from directly reverting to the operating mode before the consecutive occurrences reach a higher-level condition, thus ensuring that the switching of control modes aligns with the risk evolution process. After the equipment enters the shutdown mode, shutdown response data continues to be collected. This shutdown response data includes drive output attenuation data after the shutdown command is issued, safety rope speed attenuation data, safety rope tension fluctuation data, and braking feedback data and locking feedback data corresponding to the actual equipment configuration.

[0018] When setting shutdown safety conditions, the output of the drive mechanism should decrease to the corresponding preset threshold range within the set response time, the speed of the safety rope should decrease to the corresponding preset threshold range within the set response time, and the tension fluctuation of the safety rope should decrease to the corresponding preset threshold range within the set response time. For the braking mechanism actually configured in the equipment, the braking mechanism should remain in the braking state. For the locking mechanism actually configured in the equipment, the locking mechanism should remain in the locking state.

[0019] If the equipment does not meet the set shutdown safety conditions, the control state of limiting the continued release of the safety rope is maintained, and the equipment is prohibited from re-entering the operating state. In this way, risk assessment, shutdown execution, and post-shutdown status confirmation can form a continuous closed loop, avoiding the situation where the equipment still has residual motion after the shutdown control is issued and is not further restrained.

[0020] After the equipment finishes operation or enters a non-operation period, the data generated during this control process is written into the updated sample set. When this control process corresponds to the operating mode, warning mode, or speed limit mode, the continuous time series samples, drop risk values, and control mode are written into the updated sample set. When this control process corresponds to the shutdown mode, the continuous time series samples, drop risk values, shutdown response data, and the judgment results of setting shutdown safety conditions are written into the updated sample set.

[0021] Subsequently, based on the drop risk value, control mode, and the judgment results of setting shutdown safety conditions, the updated sample set is re-labeled and classified into normal operation samples, abnormal evolution samples, or shutdown response abnormal samples. Then, during the non-operation period of the equipment, the labeled updated sample set is used to incrementally train the neural network model to update the model parameters of the neural network model.

[0022] By continuously writing new data during actual operation and participating in model updates, the updated neural network model can continuously adapt to state changes under different operating conditions, thereby maintaining consistency between risk assessment and the actual operating state of the equipment.

[0023] The beneficial effects of this invention are as follows: 1. This invention performs unified sampling frequency processing, abnormal pulse processing, missing sampling point compensation processing, and standardization processing on safety rope status data, load-bearing body motion data, and actuator status data corresponding to the actual equipment configuration. This enables operational data from different sources to participate in subsequent analysis under a unified time reference and data scale. Furthermore, it forms continuous time-series samples according to sliding time windows, and jointly extracts load change features, motion matching features, and response consistency features within the same continuous time-series sample. This ensures that changes in safety rope force, load-bearing body motion, and actuator response are no longer judged separately, but rather form corresponding risk characterization information. This improves the ability to continuously identify the risk formation process and reduces misjudgments caused by judging based on only a single instantaneous signal.

[0024] 2. This invention inputs the drop risk characterization information corresponding to continuous time-series samples into a neural network model to obtain the drop risk value corresponding to the current time window. It then combines the risk interval and continuous count values ​​within multiple consecutive time windows to progressively determine the control mode in the order of shutdown mode, speed limit mode, warning mode, and operation mode. This ensures that the switching of control modes is based on continuous risk changes, rather than on a single abnormal trigger. At the same time, when a higher risk state has not yet reached a higher level of control conditions, the equipment can still enter an intermediate control mode corresponding to the current risk level. This gives risk control a graded transition process, thereby improving the matching between control actions and the risk evolution process and preventing high-risk states from being directly ignored or falling back to normal operation when they occur in a short period of time.

[0025] 3. This invention, by continuing to collect shutdown response data after determining the shutdown mode, and by further confirming changes in the output of the drive mechanism, changes in the movement of the safety rope, fluctuations in the tension of the safety rope, and the feedback status of the actuator corresponding to the actual configuration of the equipment, enables shutdown control to go beyond simply issuing shutdown commands. It also allows for further assessment of whether the equipment meets the set shutdown safety conditions. If the equipment does not meet the set shutdown safety conditions, the control state that restricts the continued release of the safety rope is maintained, and the equipment is prohibited from re-entering the operating state. This improves the reliability of shutdown result confirmation. At the same time, by writing the data generated in each control process into the updated sample set and incrementally training the neural network model during non-operational periods, the model can continuously absorb new samples under actual working conditions and gradually adapt to state changes in different operating scenarios, thereby improving the consistency between subsequent risk identification and on-site operating status. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the process of generating fall risk characterization information according to the present invention. Figure 2 This is a flowchart of the control mode determination and shutdown safety confirmation process of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] like Figures 1 to 2As shown, this embodiment of the invention provides a method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning, for use in work equipment; the work equipment is equipped with a safety rope, a drive mechanism, a controller, and one of a braking mechanism and a locking mechanism, or both a braking mechanism and a locking mechanism; in this embodiment, the braking mechanism status data, locking mechanism status data, braking feedback data, locking feedback data, and their corresponding stopping safety conditions are collectively referred to as the actuator-related data and actuator-related conditions corresponding to the actual configuration of the equipment.

[0029] During equipment operation, the controller collects operational data according to a set sampling period. This operational data includes safety rope status data, carrier motion data, and actuator status data. Specifically, safety rope tension data is collected by a tension sensor installed along the safety rope's force path; safety rope length variation data is collected by a length encoder installed at the rope winding / unwinding section; and safety rope speed data is collected by a speed detection unit. Carrier displacement, velocity, and acceleration data are collected by an inertial measurement unit and a displacement detection unit installed on the carrier. Actuator status data is collected by the corresponding status detection units for the drive mechanism, braking mechanism, and locking mechanism. For equipment with only a braking mechanism, drive mechanism status data and braking mechanism status data are collected; for equipment with only a locking mechanism, drive mechanism status data and locking mechanism status data are collected; for equipment with both braking and locking mechanisms, drive mechanism status data, braking mechanism status data, and locking mechanism status data are collected simultaneously. Through this data collection method, the data used in subsequent calculations directly corresponds to the actual configuration and operating state of the equipment.

[0030] After the controller obtains the operating data, it first unifies the data with different sampling frequencies to the same sampling frequency, then removes or smooths abnormal pulse data, fills in missing sampling points, and converts the data of different physical quantities into standardized data under a unified dimension. After completing the preprocessing, the data is divided according to a sliding time window of a set length to form continuous time series samples. The continuous time series samples consist of the current time window and one or more historical time windows located before the current time window.

[0031] In this embodiment, the length of the sliding time window, the number of historical time windows, and the data sliding step size are set according to the device operating speed and control response cycle, so that continuous time series samples can cover continuous state changes before and after the formation of risk.

[0032] After forming continuous time-series samples, the controller generates fall risk characterization information based on the data within the same continuous time-series sample. The fall risk characterization information includes load change characteristics, motion matching characteristics, and response consistency characteristics. Load change characteristics are generated from safety rope tension data, specifically including the safety rope tension fluctuation amplitude, safety rope tension change rate, and safety rope tension abrupt change. Motion matching characteristics are generated from time-synchronized safety rope length change data, safety rope movement speed data, load-bearing body displacement data, load-bearing body velocity data, and load-bearing body acceleration data, specifically including the deviation between the safety rope length change and the load-bearing body displacement change, the deviation between the safety rope movement speed and the load-bearing body velocity, and the deviation between the safety rope movement speed change and the load-bearing body acceleration.

[0033] The response consistency feature is generated from the control command status data, drive output status data and actuator status data corresponding to the actual equipment configuration after time synchronization. Specifically, it includes the drive output change deviation value after the control command is issued, as well as the braking action feedback deviation value and locking action feedback deviation value corresponding to the actual equipment configuration. The above three types of features correspond to the change of safety rope force, the motion relationship of the carrier and the response of the actuator, respectively, and can jointly characterize the main state changes in the risk formation process.

[0034] Subsequently, the controller inputs the fall risk characterization information into a pre-trained neural network model to obtain the fall risk value corresponding to the current time window. In this embodiment, the neural network model is trained using normal operation samples, abnormal evolution samples, and abnormal shutdown response samples. Normal operation samples correspond to operation processes where all operating data are within a preset range. Abnormal evolution samples correspond to operation processes where at least one of the following exceeds a preset range: changes in safety rope force, deviation between safety rope movement and load-bearing body movement, or lag in actuator response. Abnormal shutdown response samples correspond to operation processes where the equipment fails to meet the set shutdown safety conditions after executing shutdown control. The neural network model uses the fall risk characterization information corresponding to continuous time-series samples as input and the sample category labeling results corresponding one-to-one with the above three types of samples as output targets. After adopting this training method, the neural network model can establish a correspondence between continuous operating states and risk levels.

[0035] During the control phase, the controller determines the control mode in descending order based on the risk range to which the drop risk value belongs and the number of consecutive occurrences of the drop risk value within multiple consecutive time windows. First, it checks whether the criteria for the shutdown mode are met. If the shutdown mode criteria are not met, it checks whether the criteria for the speed limiting mode are met. If neither the shutdown nor the speed limiting mode criteria are met, it checks whether the criteria for the warning mode are met. If none of the aforementioned criteria are met, the operating mode is determined. This progressive judgment method avoids the direct reversion to the operating mode when a higher risk value occurs in a short period of time, ensuring that the switching order of the control modes is consistent with the order of risk evolution.

[0036] When the controller determines that the current control mode is shutdown mode, it first controls the drive mechanism to stop output. For equipment equipped with a braking mechanism, it controls the braking mechanism to enter the braking state; for equipment equipped with a locking mechanism, it controls the locking mechanism to enter the locking state; for equipment equipped with both a braking mechanism and a locking mechanism, it controls the braking mechanism and the locking mechanism to act synchronously to limit the safety rope from continuing to release. After executing the shutdown control, it continues to collect shutdown response data within a set response time. The shutdown response data includes drive output attenuation data after the shutdown command is issued, safety rope movement speed attenuation data, safety rope tension fluctuation data, and braking feedback data and locking feedback data corresponding to the actual equipment configuration. The controller determines whether the equipment meets the set shutdown safety conditions based on the shutdown response data.

[0037] The set shutdown safety conditions include: the output of the drive mechanism drops to a preset threshold range within a set response time, the speed of the safety rope drops to a preset threshold range within a set response time, and the tension fluctuation of the safety rope drops to a preset threshold range within a set response time; for equipment equipped with a braking mechanism, the braking mechanism is also required to maintain a braking state; for equipment equipped with a locking mechanism, the locking mechanism is also required to maintain a locking state.

[0038] If the equipment does not meet the set shutdown safety conditions, the aforementioned control state of limiting the continued release of the safety rope will be maintained, and the equipment will be prohibited from re-entering the operating state. By continuing to perform state confirmation after the shutdown control is issued, it is possible to avoid the equipment still having residual motion after the shutdown control is issued and thus not being further restrained.

[0039] After the equipment finishes operation or enters a non-operation period, the controller writes the data generated during this control process into the updated sample set. When this control process corresponds to the operating mode, warning mode, or speed limit mode, the controller writes the continuous time sequence samples, drop risk values, and control mode into the updated sample set. When this control process corresponds to the shutdown mode, the controller writes the continuous time sequence samples, drop risk values, shutdown response data, and the judgment results of setting shutdown safety conditions into the updated sample set.

[0040] Subsequently, based on the drop risk value, control mode, and the judgment results of the set shutdown safety conditions, the updated sample set is labeled with sample categories and classified into normal operation samples, abnormal evolution samples, or abnormal shutdown response samples. Finally, during the equipment's non-operational periods, the labeled updated sample set is used to incrementally train the neural network model to update the model parameters. By continuously incorporating new data from the actual operation process into the model update, the updated neural network model can continuously adapt to the state changes under actual operating conditions.

[0041] In this embodiment, the carrier is a basket, work platform or load component that is connected to the safety rope and moves with the safety rope. In order to make the subsequent risk assessment based on data corresponding to the actual operating status of the equipment, the operating data includes safety rope status data, carrier movement data and actuator status data. The safety rope status data includes safety rope tension data, safety rope length change data, and safety rope speed data, which are used to reflect the force state, length change state, and motion state of the safety rope during operation, respectively. The carrier motion data includes carrier displacement data, carrier velocity data, carrier acceleration data, and carrier posture data, which are used to reflect the position and motion changes of the carrier under the constraint of the safety rope.

[0042] The actuator status data includes drive output status data, as well as braking mechanism status data, locking mechanism status data, control command status data, and execution feedback status data corresponding to the actual configuration of the equipment. The control command status data is the control command data output by the controller to the drive mechanism and the braking and locking mechanisms corresponding to the actual configuration of the equipment. The execution feedback status data is the action feedback data of the parts of the drive mechanism, braking mechanism, and locking mechanism that correspond to the actual configuration of the equipment.

[0043] Specifically, safety rope tension data can be obtained by a tension detection unit installed along the force path of the safety rope; safety rope length change data and safety rope speed data can be obtained by a detection unit installed at the rope winding / unwinding section; carrier displacement data, carrier velocity data, carrier acceleration data, and carrier attitude data can be obtained by a motion detection unit installed on the carrier; and actuator status data can be obtained by the corresponding actuator status detection unit. For equipment equipped with a braking mechanism, braking mechanism status data is collected; for equipment equipped with a locking mechanism, locking mechanism status data is collected; and for equipment equipped with both braking and locking mechanisms, both braking and locking mechanism status data are collected simultaneously.

[0044] Through the above data settings, the status of the safety rope, the motion status of the carrier, and the response status of the actuator can form a corresponding data basis during the same operation, thereby providing basic data for the subsequent generation of fall risk characterization information.

[0045] In this embodiment, after acquiring the running data, the controller first preprocesses the running data. Specifically, in the case of inconsistent sampling frequencies of running data from different sources, the controller resamples the data with different sampling frequencies into data with a unified sampling frequency based on the timestamp or sampling sequence number corresponding to each running data, so that each running data can be processed subsequently under the same time reference. After unified sampling is completed, abnormal pulse values ​​generated by instantaneous sensor interference or signal jitter in the data sequence are removed or smoothed to reduce the impact of instantaneous outliers on subsequent risk assessment. For missing sampling points that occur during the acquisition process, the controller performs supplementary value processing based on the adjacent valid sampling values ​​before and after the missing sampling point to maintain the continuity of the data sequence. After the above processing is completed, the data corresponding to different physical quantities in the safety rope status data, carrier motion data, and actuator status data are mapped to standardized data under a unified numerical range, so that various types of operational data can participate in subsequent model calculations on the same data scale.

[0046] Through the above preprocessing, the original operational data can be kept consistent in terms of time base, data continuity and data scale, thereby providing a data foundation for subsequent time window division, continuous time series sample formation and fall risk characterization information generation.

[0047] In this embodiment, after completing the preprocessing of the running data and forming continuous time-series samples, the controller generates fall risk characterization information based on the data within the same continuous time-series sample; the fall risk characterization information includes load change characteristics, motion matching characteristics, and response consistency characteristics; wherein, the load change characteristics are generated from the safety rope tension data within the same continuous time-series sample.

[0048] Specifically, the controller calculates the changes in safety rope tension data between adjacent sampling points within the same continuous time-series sample, forming a load change characteristic value that reflects the change in force on the safety rope. The motion matching feature is generated from the time-synchronized safety rope length change data, safety rope motion speed data, load displacement data, load velocity data, and load acceleration data within the same continuous time-series sample.

[0049] Under a unified time reference, the controller compares the corresponding sampling points of the above data and calculates the corresponding differences and changes to form motion matching characteristic values ​​that reflect the degree of deviation between the movement of the safety rope and the movement of the load-bearing body. The response consistency characteristics are generated by the control command status data, drive output status data and braking mechanism status data, locking mechanism status data and actuator action feedback data that are synchronized within the same continuous time series sample.

[0050] Under a unified time reference, the controller compares the changes in drive output and actuator action feedback after the control command is issued, and calculates the corresponding difference and change to form a response consistency characteristic value that reflects the degree of deviation between the control command and the actual response of the actuator.

[0051] By incorporating the changes in the force on the safety rope, the motion deviation between the safety rope and the load-bearing body, and the response deviation of the actuator into the same continuous time-series sample for joint processing, the generated fall risk characterization information can cover the main state changes in the risk formation process and provide a corresponding data basis for the input of the subsequent neural network model.

[0052] In this embodiment, the aforementioned load change characteristics, motion matching characteristics, and response consistency characteristics are each composed of corresponding specific feature quantities.

[0053] For load variation characteristics, the controller extracts the safety rope tension fluctuation amplitude, safety rope tension change rate, and tension change between adjacent sampling points based on the safety rope tension data within the same continuous time series sample. Through the above features, the force fluctuation, force change rate, and instantaneous force change of the safety rope within the current time window can be reflected.

[0054] For motion matching features, the controller calculates the deviation between the change in safety rope length and the change in load displacement, the deviation between the safety rope speed and the load speed, and the deviation between the change in safety rope speed and the load acceleration based on the time-synchronized safety rope length change data, safety rope speed data, load displacement data, load speed data, and load acceleration data.

[0055] The aforementioned deviation values ​​reflect the degree of matching between the movement of the safety rope and the movement of the load-bearing body. For the response consistency characteristic, the controller calculates the drive output change deviation value based on the change in drive output after the control command is issued. For equipment equipped with a braking mechanism, the braking action feedback deviation value is calculated. For equipment equipped with a locking mechanism, the locking action feedback deviation value is calculated. These deviation values ​​reflect the degree of deviation between the actual response of the actuator and the target response corresponding to the control command after the control command is issued. The aforementioned specific feature quantities together constitute the fall risk characterization information and serve as input data for the subsequent neural network model.

[0056] In this embodiment, the training samples of the neural network model are obtained by organizing continuous time-series samples formed during the historical operation of the device. When classifying the training samples, it is first determined whether the operation process corresponding to the sample has executed shutdown control and has not met the shutdown safety conditions set above. If the condition is met, the sample is identified as a shutdown response abnormal sample.

[0057] For the remaining samples that were not identified as abnormal shutdown responses, they are further classified according to whether the status of the safety rope, the movement status of the carrier, and the status of the actuator corresponding to the actual configuration of the equipment are within the corresponding preset threshold range.

[0058] When the state of the safety rope, the motion state of the carrier, and the state of the actuator corresponding to the actual configuration of the equipment are all within the corresponding preset threshold range, the sample is determined as a normal operation sample; when at least one of the following exceeds the corresponding preset threshold range: the change in force on the safety rope, the deviation between the movement of the safety rope and the movement of the carrier, and the response hysteresis of the actuator, the sample is determined as an abnormal evolution sample.

[0059] Using the above classification method, the training samples are divided into normal operation samples, abnormal evolution samples, and shutdown response abnormal samples, which correspond to stable operation state, risk evolution state, and shutdown response abnormal state, respectively, and serve as the three types of samples used for training the neural network model.

[0060] In this embodiment, after classifying normal operation samples, abnormal evolution samples, and shutdown response abnormal samples, the neural network model is trained using the above three types of samples.

[0061] Specifically, the controller first extracts the corresponding fall risk characterization information from each continuous time series sample and uses the fall risk characterization information as input data for the neural network model; at the same time, it assigns corresponding sample category labeling results to normal operation samples, abnormal evolution samples and shutdown response abnormal samples, and uses the sample category labeling results as output labeling data during neural network model training.

[0062] During training, the controller iteratively adjusts the model parameters of the neural network model based on the deviation between the output of the neural network model and the corresponding sample category labeling results, so that the output of the neural network model gradually approaches the corresponding sample category labeling results.

[0063] After the above training, the neural network model can distinguish the feature differences between stable operation state, risk evolution state and abnormal shutdown response state, and serve as the model basis for subsequent risk identification based on current continuous time series samples.

[0064] In this embodiment, the controller divides the fall risk value into a first interval, a second interval, and a third interval, which are progressively more risky and do not overlap, based on a preset threshold for the fall risk value.

[0065] When determining the control mode, the controller not only determines which risk range the fall risk value corresponding to the current time window falls into, but also counts the number of times the fall risk value falls into the corresponding risk range in multiple consecutive time windows.

[0066] To ensure that the switching order of control modes is consistent with the changing order of risk levels, the first set number of times is less than or equal to the second set number of times, and the second set number of times is less than or equal to the third set number of times.

[0067] In specific judgments, the controller makes judgments in the following order: shutdown mode, speed limit mode, warning mode, and operation mode.

[0068] First, determine whether the fall risk value falls into the third interval, and whether the count of consecutive fall risk values ​​falling into the third interval has reached the third set number; if so, determine that the current control mode is the shutdown mode.

[0069] If the conditions for determining the shutdown mode are not met, it is further determined whether the drop risk value falls into the range of not less than the second interval, and whether the count value of the drop risk value falling into the range of not less than the second interval for a continuous period of time reaches the second set number; if it does, the current control mode is determined to be the speed limit mode.

[0070] If the speed limit mode is still not met, the system continues to determine whether the fall risk value falls into the range of the first interval and whether the count of consecutive fall risk values ​​falling into the range of the first interval reaches the first set number; if so, the current control mode is determined to be the warning mode.

[0071] If none of the above judgment conditions are met, the controller determines the current control mode as the operating mode. Through the above progressive judgment method, a higher risk state will still enter the intermediate control mode corresponding to the current risk level if the higher level control conditions are not met.

[0072] In this embodiment, after the controller determines that the current control mode is the shutdown mode and issues a shutdown command, it continues to collect shutdown response data within the set response time.

[0073] The shutdown response data includes drive output attenuation data after the shutdown command is issued, safety rope movement speed attenuation data, safety rope tension fluctuation data, as well as braking feedback data and locking feedback data corresponding to the actual equipment configuration.

[0074] Among them, the drive output attenuation data is used to characterize the output change process of the drive mechanism after the stop command is issued, the safety rope motion speed attenuation data is used to characterize the change process of the safety rope from the motion state to the stop state, and the safety rope tension fluctuation data is used to characterize the change of force on the safety rope during the stop process.

[0075] For equipment equipped with a braking mechanism, braking feedback data is further collected; for equipment equipped with a locking mechanism, locking feedback data is further collected; for equipment equipped with both a braking mechanism and a locking mechanism, both braking feedback data and locking feedback data are collected simultaneously.

[0076] The controller determines whether the equipment meets the set shutdown safety conditions based on the above shutdown response data; specifically, the output of the drive mechanism should decrease to its corresponding preset threshold range within the set response time, the speed of the safety rope should decrease to its corresponding preset threshold range within the set response time, and the tension fluctuation of the safety rope should decrease to its corresponding preset threshold range within the set response time.

[0077] For equipment equipped with a braking mechanism, the braking mechanism must be kept in a braking state; for equipment equipped with a locking mechanism, the locking mechanism must be kept in a locking state. If any of the above conditions corresponding to the actual configuration of the equipment are not met, the equipment is deemed not to meet the set shutdown safety conditions.

[0078] By continuing to confirm the changes in drive output, safety rope movement, and actuator feedback status after the shutdown command is issued, the shutdown result can be confirmed, and it can be determined whether the equipment has entered a controlled shutdown state.

[0079] In this embodiment, after each control process ends, the controller writes the data generated during the current control process into the updated sample set. For the operating mode, warning mode or speed limit mode corresponding to the current control process, the controller writes the continuous time sequence sample, drop risk value and control mode into the updated sample set.

[0080] For the shutdown mode corresponding to this control process, the controller, based on writing continuous timing samples, drop risk values ​​and control modes, further writes shutdown response data and judgment results of setting shutdown safety conditions into the updated sample set. In this way, the data generated under different control modes can be written into the updated sample set with content corresponding to their operation process.

[0081] After data writing is complete, the controller labels the newly added samples in the updated sample set. Specifically, the controller first determines whether the control mode corresponding to the newly added sample is a shutdown mode and whether the equipment fails to meet the set shutdown safety conditions. If both conditions are met, the newly added sample is labeled as a shutdown response anomaly sample. For the remaining newly added samples not labeled as shutdown response anomalies, the controller further classifies them based on whether the safety rope status, carrier movement status, and actuator status corresponding to the actual equipment configuration are within the corresponding preset threshold ranges. When the safety rope status, carrier movement status, and actuator status corresponding to the actual equipment configuration are all within the corresponding preset threshold ranges, the newly added sample is labeled as a normal operation sample. When at least one of the above states exceeds the corresponding preset threshold range, the newly added sample is labeled as an abnormal evolution sample. Through the above labeling order, the classification rules for newly added samples can be kept consistent with the classification rules for the aforementioned training samples, and the same sample can be avoided from being cross-classified between shutdown response anomalies and abnormal evolution samples.

[0082] After the equipment finishes operation or enters a non-operational period, the controller extracts new samples with completed sample category labeling from the updated sample set and inputs these new samples as incremental training samples into the neural network model. During the training process, the controller continues to adjust the model parameters of the neural network model based on the fall risk characterization information and sample category labeling results corresponding to the incremental training samples. After incremental training, the neural network model can be updated based on the data generated during the latest operation of the equipment, so that the neural network model can continuously adapt to the state changes under actual operating conditions.

[0083] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0084] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting, warning, and automatically shutting down safety rope fall risks based on machine learning, characterized by: The method, applicable to work equipment equipped with one or more of a safety rope, a drive mechanism, a controller, a braking mechanism, and a locking mechanism, comprises: Collect operational data, including safety rope status data, carrier motion data, and actuator status data; The running data is preprocessed and divided according to a sliding time window of a set length to obtain a continuous time series sample consisting of the current time window and one or more historical time windows preceding the current time window; Fall risk characterization information is generated based on the continuous time-series samples. The fall risk characterization information includes load variation characteristics, motion matching characteristics, and response consistency characteristics. The fall risk representation information is input into a pre-trained neural network model, which outputs a fall risk value corresponding to the current time window. Based on the risk range to which the fall risk value belongs and the number of times the fall risk value appears consecutively in multiple consecutive time windows, the corresponding control mode is determined, and the control signal corresponding to the control mode is output. When the control mode is the shutdown mode, the controller controls the drive mechanism to stop outputting and controls one or both of the braking mechanism and the locking mechanism to enter a control state that restricts the continued release of the safety rope. After executing the shutdown control, shutdown response data is collected within the set response time, and the shutdown response data is used to determine whether the equipment meets the set shutdown safety conditions. If the equipment fails to meet the set shutdown safety conditions, the controller maintains the control state of limiting the release of the safety rope and prohibits the equipment from re-entering the operating state.

2. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 1, characterized in that: The carrier is a suspended basket, work platform, or load component connected to the safety rope and moving with the safety rope; the safety rope status data includes safety rope tension data, safety rope length change data, and safety rope movement speed data; the carrier motion data includes carrier displacement data, carrier velocity data, carrier acceleration data, and carrier attitude data; the actuator status data includes drive output status data, braking status data, locking status data, control command status data, and execution feedback status data.

3. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 2, characterized in that: The preprocessing of the running data includes: converting data with different sampling frequencies into data with a unified sampling frequency, removing or smoothing abnormal pulse data, filling in missing sampling points, and converting data of different physical quantities into standardized data for model calculation.

4. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 3, characterized in that: The load variation feature is generated based on the safety rope tension data within the same continuous time series sample; the motion matching feature is generated based on the difference and change of corresponding sampling points of the safety rope length variation data, safety rope motion speed data, carrier displacement data, carrier velocity data, and carrier acceleration data after time synchronization within the same continuous time series sample; the response consistency feature is generated based on the difference and change of corresponding sampling points of the control command status data, drive output status data, braking status data, locking status data, and execution feedback status data after time synchronization within the same continuous time series sample.

5. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 4, characterized in that: The load variation characteristics include the amplitude of safety rope tension fluctuation, the rate of change of safety rope tension, and the sudden change in safety rope tension; the motion matching characteristics include the deviation between the change in safety rope length and the change in load displacement, the deviation between the speed of the safety rope and the speed of the load, and the deviation between the change in the speed of the safety rope and the acceleration of the load; the response consistency characteristics include the deviation of the drive output change after the control command is issued, the deviation of the braking action feedback, and the deviation of the locking action feedback.

6. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 5, characterized in that: The training samples used to train the neural network model include normal operation samples, abnormal evolution samples, and abnormal shutdown response samples. The normal operation samples correspond to the operation process in which the state of the safety rope, the motion state of the carrier, and the state of the actuator are all within their respective preset threshold ranges. The abnormal evolution samples correspond to the operation process in which at least one of the following exceeds its respective preset threshold range: change in the force of the safety rope, deviation between the movement of the safety rope and the movement of the carrier, or lag in the response of the actuator. The abnormal shutdown response samples correspond to the operation process in which the equipment fails to meet the set shutdown safety conditions after the shutdown control is executed.

7. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 6, characterized in that: When training the neural network model using the normal operation samples, the abnormal evolution samples, and the shutdown response abnormal samples, the fall risk representation information corresponding to the continuous time series samples is used as the model input, and the first label, the second label, and the third label, which correspond one-to-one with the normal operation samples, the abnormal evolution samples, and the shutdown response abnormal samples, are used as the model output targets.

8. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 7, characterized in that: The risk range includes a first range, a second range, and a third range, with the risk level increasing sequentially from the first range to the third range. The corresponding control mode is determined based on the risk range to which the fall risk value belongs and the number of consecutive occurrences of the fall risk value within multiple consecutive time windows, including sequentially judging: When the drop risk value falls into the third range and reaches the third set number of times consecutively, it is determined to be in shutdown mode; When the conditions for determining the shutdown mode are not met, if the drop risk value falls into the second interval or the third interval and reaches the second set number of times consecutively, it is determined to be a speed limit mode. When the conditions for determining the shutdown mode and the conditions for determining the speed limit mode are not met, if the drop risk value falls into the first interval, the second interval, or the third interval and reaches the first set number of times consecutively, it is determined to be a warning mode. If none of the aforementioned criteria are met, the system is determined to be in operating mode. Wherein, the first set number of times is less than or equal to the second set number of times, and the second set number of times is less than or equal to the third set number of times.

9. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 8, characterized in that: The shutdown response data includes drive output attenuation data after the shutdown command is issued, safety rope movement speed attenuation data, safety rope tension fluctuation data, braking feedback data, and locking feedback data. The set shutdown safety conditions include: the output of the drive mechanism drops to a corresponding preset threshold range within the set response time; the speed of the safety rope drops to a corresponding preset threshold range within the set response time; the tension fluctuation of the safety rope drops to a corresponding preset threshold range within the set response time; the braking mechanism remains in a braking state; and the locking mechanism remains in a locking state. If any one of these conditions is not met, the equipment is deemed not to have met the set shutdown safety conditions.

10. The method for predicting, warning, and automatically stopping safety rope fall risks based on machine learning according to claim 9, characterized in that: When the control mode corresponding to this control process is the operation mode, the warning mode or the speed limit mode, the continuous time series sample, the drop risk value and the control mode are written into the update sample set. When the control mode corresponding to this control process is the shutdown mode, the continuous time series sample, the drop risk value, the control mode, the shutdown response data, and the judgment result of the set shutdown safety conditions are written into the update sample set; Based on the judgment results of the drop risk value, the control mode, and the set shutdown safety conditions, the updated sample set is labeled as the normal operation sample, the abnormal evolution sample, or the shutdown response abnormal sample. During the non-operation period of the equipment, the neural network model is incrementally trained based on the labeled updated sample set to update the model parameters of the neural network model.

Images