Automatic elevator interlock control method based on travel switch

Abstract

The present application belongs to the technical field of control, and particularly relates to an automatic hoist interlocking control method based on a travel switch, which comprises the following steps: synchronously collecting switch level sequences and hoist electromechanical operation parameters; establishing a model based on historical data, predicting a trigger time window and a reference feature template of a current working condition; when a jump is monitored in the window, extracting a transient level feature of the jump start to form a measured feature vector; calculating the similarity of the measured feature and the reference template to obtain a confidence degree; setting a dynamic threshold based on model error evaluation; and when the confidence degree is higher than the threshold, confirming effective switching and triggering an interlocking instruction. The present application can effectively filter out false trigger signals caused by mechanical vibration or electrical interference, while ensuring the rapidity of system response, and improving the accuracy and overall safety of the interlocking control system.

Description

Technical Field

[0001] This invention relates to the field of control technology. More specifically, this invention relates to an interlocking control method for automatic hoists based on limit switches. Background Technology

[0002] Interlocking control systems primarily rely on limit switches installed on the derrick or track to sense the position of the hoisting container and trigger corresponding control logic. In practice, the mechanical contacts of the limit switches inevitably experience signal jitter at the moment of closing or opening due to elastic collisions and arcing. That is, before stabilizing to a new state, the output level will undergo a series of brief and irregular jumps between high and low levels. If the control system directly adopts this unprocessed jump signal, it is highly susceptible to misjudgment of the state, leading to control logic disorder and hoist malfunction.

[0003] Currently, conventional debouncing methods mainly include hardware and software approaches. Hardware methods typically employ RC filter circuits, utilizing the charging and discharging characteristics of capacitors to smooth signal transitions. However, once the circuit parameters are set, they remain fixed, making it difficult to adapt to the dynamic changes in the jitter characteristics of limit switches during long-term use. Software methods, on the other hand, often use fixed-delay debouncing algorithms. This means that after detecting the first level transition, the program forcibly waits for a fixed delay, and only after the estimated jitter period has ended is the level sampled again to confirm a stable state.

[0004] However, traditional fixed-delay software methods have significant limitations in complex real-world operating conditions. On one hand, if the delay time is set too short, it may fail to effectively filter out prolonged, severe jitter, leading to de-jitter failure; if set too long, it inevitably introduces unnecessary system response delays, reducing the real-time performance of interlocking control. Especially in situations where the hoist is operating at high speed or requires emergency safety response, excessively long delays may delay critical safety protection actions. On the other hand, the actual jitter characteristics of limit switches are not static but change with the aging and wear of the switches themselves, variations in the working environment, and the real-time operating conditions of the hoist, such as differences in load and speed. Therefore, existing de-jitter methods cannot flexibly adjust the identification strategy according to the actual operating conditions of the hoist and the dynamic changes in the switching components, failing to effectively balance the reliability of signal de-jittering with the speed of system response. How to dynamically and accurately identify the actual switch trigger signal by combining multi-dimensional state parameters during hoist operation, thereby improving the accuracy and safety of the automatic interlocking control system, has become a pressing technical bottleneck in the field of hoist control. Summary of the Invention

[0005] To address the technical problems of existing de-jittering methods being unable to dynamically adjust according to operating conditions and struggling to balance response speed and reliability, this invention proposes an automatic hoist interlocking control method based on limit switches. This method can dynamically predict trigger time by combining electromechanical operating parameters and accurately identify switch status through multi-dimensional feature matching and dynamic threshold determination. While adapting to changing operating conditions, it improves the safety and judgment accuracy of the interlocking control system.

[0006] This invention provides an interlocking control method for automatic hoists based on limit switches, comprising: synchronously acquiring high-frequency level sequences of limit switches and electromechanical operating parameters during hoist operation; establishing a prediction model based on historical electromechanical operating parameters and confirmed valid limit switch level jitter data, wherein the prediction model predicts online the expected trigger time window of the limit switch matching the current operating condition based on the currently acquired electromechanical operating parameters, and retrieves or generates a reference jitter feature template representing typical level jitter characteristics under the current operating condition; within the expected trigger time window, when a level jump of the limit switch is detected, extracting the transient level feature at the start of the jump to form a measured jitter feature vector; calculating the similarity between the measured jitter feature vector and the reference jitter feature template to obtain an initial confidence level representing the degree of matching between the actual jitter and the expected jitter; setting a decision threshold based on historical prediction error statistics or real-time uncertainty assessment of the prediction model; and confirming a valid limit switch state switch switching when the initial confidence level is higher than the decision threshold, and triggering the corresponding interlocking control command.

[0007] This invention enhances the adaptability and accuracy of signal judgment by synchronously collecting multi-dimensional electromechanical operating parameters and establishing a prediction model, comparing the real-time jitter characteristics with the expected template, confirming the switch status by combining dynamically set decision thresholds, and realizing online model updates.

[0008] Preferably, the establishment of the prediction model includes: the prediction model is a state-space model, which takes the electromechanical operating parameters of the hoist as input, and takes the start and end points of the expected trigger time window and the reference jitter feature template as model output; the initial structure and parameters of the model are determined offline by an identification method, and the parameters are recursively corrected using new samples during online operation.

[0009] This invention constructs a state-space model, uses offline identification to determine initial parameters, and recursively corrects them using new samples during runtime, enabling the model to continuously adapt to the dynamic changes of the system.

[0010] Preferably, the step of extracting the transient level features at the start of the transition to form a measured jitter feature vector includes: taking the first level transition as the starting point, extracting a very short transient level sequence, such as 5ms, as an analysis segment; extracting time-domain features from the analysis segment, including the transition slope, transient pulse width, and effective level time ratio; and combining these time-domain feature parameters to form the measured jitter feature vector.

[0011] This invention extracts time-domain feature parameters from the truncated extremely short level sequence to form a measured vector, which accurately reflects the transient physical characteristics of the jitter signal while ensuring the system's extremely fast response.

[0012] Preferably, the step of calculating the similarity between the measured jitter feature vector and the reference jitter feature template to obtain an initial confidence level representing the degree of matching between the actual jitter and the expected jitter includes: treating the measured jitter feature vector and the reference jitter feature template as two points in a multidimensional feature space, and calculating the weighted Euclidean distance or Mahalanobis distance between the two points; converting the calculated distance value into a similarity value between 0 and 1 through a preset mapping function, and using it as the initial confidence level.

[0013] This invention calculates the weighted distance between measured features and a reference template, and uses a mapping function to convert it into an intuitive similarity value, thus reasonably evaluating the degree of matching between the actual signal and the expected signal.

[0014] Preferably, the step of statistically analyzing historical prediction errors or assessing real-time uncertainty based on the prediction model and setting a decision threshold includes: tracking and statistically analyzing the deviation between the predicted value of the prediction model for the center point of the expected trigger time window and the actual confirmed trigger time point in real time, calculating the standard deviation in the most recent time period; normalizing the standard deviation by dividing it by a preset reference time value, multiplying it by a preset risk coefficient, subtracting the multiplication result from a fixed basic threshold between 0 and 1, and taking the minimum threshold if the result is less than the preset minimum threshold. The result is the final decision threshold under the current working condition.

[0015] This invention dynamically calculates the final decision threshold under the current working condition by statistically analyzing the standard deviation of the prediction error in real time and combining it with preset parameters such as the risk coefficient. It can automatically adjust the judgment criteria when the uncertainty of the model prediction changes.

[0016] Preferably, it further includes: using the confirmed effective measured jitter feature vector as a new sample to update the prediction model online; specifically, it includes: using a weighted average strategy to update the reference jitter feature template that matches the current operating condition using the measured jitter feature vector.

[0017] This invention employs a weighted average strategy to update the reference jitter feature template that is most closely clustered with the current operating condition, enabling the reference template to continuously track the gradual change process of limit switch jitter characteristics.

[0018] Preferably, the step of confirming a valid limit switch state switch when the initial confidence level is higher than the decision threshold and triggering the corresponding interlock control command includes: after confirming a valid limit switch state switch, generating a jitter-free and logically deterministic switch state signal; sending the switch state signal to the hoist safety control logic unit, which, according to a preset safety procedure, coordinates and executes the actions of cutting off the drive motor power and engaging the braking system in a predetermined sequence to achieve reliable stopping of the hoist.

[0019] After confirming a valid switch, the present invention outputs a logically determined switch status signal, and the safety control logic unit coordinates the actions of the motor and brake according to a predetermined sequence, thereby achieving a safe and smooth stop of the hoist.

[0020] Preferably, the synchronous acquisition of the high-frequency level sequence of the hoist limit switch and the electromechanical operating parameters during the hoist operation includes: acquiring the binary high and low level states of the limit switch through a digital input channel to form a high-frequency level sequence; simultaneously, synchronously reading electromechanical operating parameters including one or more combinations of speed, position, drive motor current or torque through a communication bus or analog input channel; all channel acquisition operations are triggered by a unified hardware clock signal.

[0021] Preferably, the step of establishing a prediction model based on historical electromechanical operating parameters and confirmed valid limit switch level jitter data includes: using an unsupervised learning algorithm to perform cluster analysis on a large dataset of electromechanical operating parameters stored in a historical database, dividing the hoist's operating state into several typical working conditions; for each working condition category, constructing and training a long short-term memory network, the input of which is the time series of electromechanical operating parameters of the hoist under a specific working condition, and the output is the remaining time from the current moment until the hoist arrives and triggers the limit switch; extracting the time-frequency feature vectors of all triggering events under the same working condition category from the historical data, performing weighted average calculation, obtaining a reference jitter feature template specific to that working condition, and storing it in a template library.

[0022] Preferably, the method further includes: using the confirmed valid measured jitter feature vector as a new sample to update the prediction model online; specifically, this includes: using the confirmed valid measured jitter feature vector, the electromechanical parameter sequence before triggering, and the actual triggering time difference as a new high-quality sample pair; storing the new high-quality sample pair in an online learning buffer; and when the number of samples in the buffer reaches a preset batch size, performing a small batch of online training on the corresponding prediction model to fine-tune the network weights, so as to achieve continuous optimization of the prediction model.

[0023] This invention achieves fine-tuning of network weights and continuous optimization of the model by storing newly confirmed high-quality samples in an online learning buffer and performing online mini-batch training on the prediction model when a preset batch is reached.

[0024] The beneficial effects of this invention are as follows: This invention synchronously acquires the electromechanical operating parameters of the hoist and the level signals of the limit switches, and uses historical data to build a predictive model to predict the reasonable triggering time range and typical jitter patterns under specific operating conditions. Within the expected time window, the real-time monitored level jitter characteristics are compared with a reference template, and a comprehensive judgment is made based on a confidence standard dynamically adjusted according to the model performance. This effectively identifies the validity of the switch signals and filters out false triggering signals caused by mechanical vibration or electrical interference. Compared with traditional fixed long-delay software jitter removal, this invention extracts only transient time-domain features within a very short time for matching and judgment, breaking the bottleneck of long-delay waiting, balancing the reliability of jitter removal with the speed of system safety protection response, and improving the accuracy of interlocking control command triggering and the overall safety of the system.

[0025] Furthermore, this invention introduces an online learning mechanism, which continuously optimizes the prediction model and reference feature template using constantly confirmed valid signals. This allows the system's judgment logic to flexibly adjust to changes in equipment operating status and the aging and wear of the limit switches themselves, thereby maintaining the stability of the hoist system and the reliability of signal recognition in long-term practical applications and under varying working conditions. Attached Figure Description

[0026] Figure 1 This is a flowchart of the automatic hoist interlocking control method based on limit switches in this invention; Figure 2 This is a schematic diagram of the hardware device in this invention. Detailed Implementation

[0027] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0028] This invention discloses an interlocking control method for automatic hoists based on limit switches, referring to... Figure 1 and Figure 2 This includes steps S1-S6: S1. Synchronously acquire the high-frequency level sequence of the limit switch of the hoist and the electromechanical operating parameters during the operation of the hoist.

[0029] In an optional embodiment, an industrial data acquisition card with multi-channel synchronous acquisition capabilities, such as the NI PCIe-6363, is used to acquire the binary high and low level states of the hoist limit switches at a sampling rate of 10kHz or higher via digital input channels, thereby forming a high-frequency level sequence. Simultaneously, existing electromechanical operating parameters, including one or more combinations of speed, position, drive motor current, or torque, are synchronously read directly from the existing hoist drive control unit or programmable logic controller (PLC) via an industrial communication bus. All channel acquisition operations are triggered by a unified hardware clock signal to ensure precise timestamp alignment of each data stream. The acquired data streams are packaged and then transferred at high speed to the industrial control computer's memory buffer via direct memory access (DMA).

[0030] S2. Based on historical electromechanical operating parameters and confirmed valid limit switch level jitter data, a prediction model is established. This prediction model predicts the expected trigger time window of the limit switch that matches the current operating condition online according to the currently collected electromechanical operating parameters, and retrieves or generates a reference jitter feature template that represents the typical level jitter characteristics under the current operating condition.

[0031] In an optional embodiment, unsupervised learning algorithms such as K-means clustering (e.g., K-Means) or Gaussian mixture models (GMM) are used to perform cluster analysis on a large dataset of electromechanical operating parameters stored in a historical database. This categorizes the hoist's operating state into several typical conditions, such as the hoist operating at high speed under no-load conditions, high speed under full load conditions, and low-speed maintenance conditions. For each condition category, a Long Short-Term Memory (LSTM) network is constructed and trained using TensorFlow or PyTorch frameworks. The input to this network is the time series of electromechanical operating parameters of the hoist under a specific condition, and the output is the remaining time until the hoist arrives and triggers the limit switch. For each triggering event in the historical data, either system-generated or manually labeled, a high-frequency level sequence within a very short time after the trigger point (e.g., 5ms) is extracted. A time-domain feature analysis algorithm is used to extract jitter features to construct a one-dimensional feature vector. The feature vectors of all triggering events within the same condition category are averaged to obtain a reference jitter feature template specific to that condition, which is then stored in a template library.

[0032] As another alternative embodiment, the prediction model can also adopt a state-space model, which takes the electromechanical operating parameters of the hoist as input, and takes the start and end points of the expected trigger time window and the reference jitter feature template as the model output; the initial structure and parameters of the model are determined offline by the identification method, and the parameters are recursively corrected using new samples during online operation.

[0033] Specifically, the input vector of the prediction model At each sampling time The speed of the hoist is collected in real time. Car position The current of phase A, phase B, and phase C of the drive motor are... , , With torque The specific calculation method for its input vector is as follows: The predicted output vector of the prediction model This includes the prediction target, that is ,in and The start and end times of the expected trigger time window. This serves as the reference jitter feature template. To characterize the system dynamics, a hidden state vector is introduced. Its state transition equation is The output equation is , where the function and The prediction model can be constructed by approximation through neural networks or nonlinear mappings.

[0034] In the offline identification phase of the predictive model, historical data covering various operating conditions of the hoist are collected. Through manual verification or post-event analysis, the trigger time point and jitter characteristics of each limit switch's effective triggering event are labeled. The electromechanical operating parameter sequence with a preset duration before triggering is used as the input sample for the predictive model, and the corresponding labeled trigger time window and jitter characteristics are used as the output label for the predictive model, thus constructing the input and output sample datasets for offline training. Then, nonlinear system identification methods, such as neural network-based system identification methods, are used to learn the dynamic structure of the model from the input and output sample datasets, completing the model initialization.

[0035] During online operation, each time a limit switch trigger is confirmed as valid, the electromechanical operating parameters of that event and the confirmed jitter feature vector constitute a new valid sample pair. Using this new sample, model parameters are augmented into the state vector, and online estimation and updates are performed using either the Extended Kalman Filter (EKF) algorithm or the Unscented Kalman Filter (UKF) algorithm. The process noise covariance matrix Q can be set as a diagonal matrix, while the measurement noise covariance matrix R is set according to the sensor accuracy. This allows the model to continuously track the slow changes in the hoist's operating characteristics caused by factors such as mechanical wear and lubrication changes, thereby achieving accurate prediction.

[0036] S3. Within the expected trigger time window, when a level change of the limit switch is detected, the transient level feature at the beginning of the change is extracted to form the measured jitter feature vector.

[0037] In an optional embodiment, during the real-time monitoring phase, the current electromechanical operating parameters are input into a trained Gaussian mixture model for condition identification. Based on the identified condition, the corresponding LSTM model is invoked, continuously inputting the latest electromechanical parameter sequence to predict the expected trigger time point T online. The expected trigger time window is set to a time range of one standard deviation before and after time T, where the standard deviation is calculated based on the historical prediction error statistics of the LSTM model. Within this expected trigger time window, edge detection is achieved by monitoring changes in the level values ​​of the digital input ports. Once the first logical transition from 0 to 1 or from 1 to 0 is detected, a transient high-frequency level sequence segment within a very short time after the transition point, such as 5ms, is immediately extracted from the memory buffer. Subsequently, the time-domain feature extraction algorithm module, identical to that used in offline processing, is invoked to process the level sequence segment and generate a measured jitter feature vector.

[0038] Specifically, transient time-domain analysis is used to extract features to form a measured jitter feature vector. That is, starting from the first level transition, a very short level sequence is extracted as an analysis segment. Time-domain features that can reflect the real physical collision characteristics are extracted from this analysis segment, including but not limited to the transition slope, the width of the first transient pulse, and the proportion of effective level time within a preset short window.

[0039] Next, the level signal of the limit switch is acquired at a sampling rate of not less than 10kHz. When the first transition from logic low to logic high is detected, the timestamp of this transition point is recorded as follows. ,by Extract a fixed-length level sequence segment from the starting point, for example, a segment of 5ms in length; that is, extract the interval as follows. The data sequence contains voltage levels over a given time period. The high / low level switching rate, the width of the first pulse lasting longer than 1 ms, and the percentage of integrated energy are calculated directly from this data sequence.

[0040] Finally, the above time-domain characteristic parameters are combined into a feature vector, which is the measured jitter feature vector.

[0041] S4. Calculate the similarity between the measured jitter feature vector and the reference jitter feature template to obtain the initial confidence level representing the degree of matching between the actual jitter and the expected jitter.

[0042] In an optional embodiment, the measured jitter feature vector and the reference jitter feature template for the corresponding operating condition retrieved from the template library are flattened into two one-dimensional long vectors. The cosine similarity function from the Scikit-learn library is used to calculate the cosine similarity between the two one-dimensional vectors. The result is a scalar value between 0 and 1, which is defined as the initial confidence level. The closer the value is to 1, the better the measured level jitter pattern matches the expected jitter pattern under the operating condition.

[0043] In another preferred embodiment, the similarity between the measured jitter feature vector and the reference jitter feature template is calculated to obtain the initial confidence level. That is, the measured jitter feature vector and the reference jitter feature template are regarded as two points in the multidimensional feature space, and the weighted Euclidean distance or Mahalanobis distance between the two points is calculated. The calculated distance value is converted into a similarity value between 0 and 1 through a preset mapping function, and this value is used as the initial confidence level.

[0044] Assume the measured jitter feature vector is The reference jitter feature template is This invention preferably uses Mahalanobis distance for calculation because it can incorporate the correlation between features. Then, based on historical valid jitter sample data, the following is calculated: Covariance matrix of eigenvectors Mahalanobis distance The calculation formula is: ,in It is the inverse of the covariance matrix.

[0045] Calculate the Mahalanobis distance Then, the Mahalanobis distance is mapped using a mapping function. Convert to range Initial confidence levels between A preferred mapping function is a Gaussian radial basis function, for example: ,in This is a scale parameter used to control the sensitivity of the distance-to-similarity conversion. Alternatively, an alternative function is... ,in The adjustment coefficient is positive.

[0046] S5. Based on the historical prediction error statistics or real-time uncertainty assessment of the prediction model, set a decision threshold.

[0047] In an optional embodiment, the decision threshold needs to be dynamically adjusted based on the current operating conditions and model performance. A moving average and standard deviation of the initial confidence levels of historical valid trigger events are maintained for each operating condition. The decision threshold is set to the moving average confidence level for that operating condition minus 1.5 times the moving standard deviation. When the model's prediction uncertainty increases, the decision threshold is correspondingly reduced to accommodate the performance degradation of the model under unfamiliar operating conditions, thereby prioritizing the preservation of trigger signals.

[0048] Specifically, a decision threshold is set based on the historical prediction error statistics of the prediction model or the real-time uncertainty assessment. That is, the deviation between the prediction value of the prediction model for the center point of the expected trigger time window and the actual confirmed trigger time point is tracked and statistically analyzed in real time, and the standard deviation in the most recent time period is calculated. After normalizing the standard deviation by dividing it by a preset reference time value, it is multiplied by a preset risk coefficient. The result of the multiplication is subtracted from a fixed basic threshold between 0 and 1. If the result is less than the preset minimum threshold, the minimum threshold is taken. The result is the final decision threshold under the current working condition.

[0049] In terms of implementation details, a first-in-first-out (FIFO) queue of length N is maintained. This queue is used to store the prediction errors of the most recent N confirmed valid trigger events. Its calculation method is as follows ,in It is the center point of the time window predicted by the model. This is the actual confirmed trigger time. The standard deviation of the N error values ​​in the queue is then calculated in real time. .

[0050] Finally, according to the formula Calculate the final threshold. Wherein, The base threshold is a fixed, relatively high value, which is preferably set to 0.9 in this invention; This is a risk coefficient used to adjust the degree of impact of uncertainty on the threshold; it is preferably set to 0.5. This is a reference time value used to normalize the standard deviation; it is preferably set to 50ms. The minimum threshold is set to prevent false positives caused by excessively low thresholds; a value of 0.65 is preferred. When the uncertainty of model prediction increases, the stringency of jitter pattern matching is appropriately reduced to prioritize ensuring that limit switch actions are not missed, thereby effectively improving the overall system availability and security.

[0051] S6. When the initial confidence level is higher than the decision threshold, it is confirmed as a valid limit switch state switch change and the corresponding interlock control command is triggered.

[0052] In an optional embodiment, the calculated initial confidence level is compared with a real-time generated decision threshold. If the initial confidence level is greater than the decision threshold, the level transition is determined to be a valid limit switch action. The system immediately sends a high-priority interlocking control command to the programmable logic controller (PLC) via the control bus, such as executing a command to disconnect the main drive motor contactor and engage the brake. After confirming the validity of the switch action, the actual trigger time is compared with the prediction time of the LSTM model or state-space model in the prediction model to calculate the prediction error. Optionally, this method further includes an online model update step. Specifically, the measured jitter feature vector (i.e., the aforementioned extracted transient time-domain feature vector), along with the electromechanical parameter sequence before triggering and the actual triggering time difference, are used as a new high-quality sample pair. Using the exponential moving average method with a small learning rate, such as 0.01, the new measured jitter feature vector is integrated into and its corresponding reference jitter feature template is updated. Simultaneously, the new sample pair is stored in an online learning buffer. When the number of samples in the buffer reaches a preset batch size, such as 32, a small batch of online training is performed on the corresponding LSTM model. Specifically, the network weights can be fine-tuned by calling the model's batch training method, i.e., the train-on-batch method, thereby achieving continuous optimization of the prediction model.

[0053] Specifically, the confirmed effective measured jitter feature vectors are used as new samples to update the prediction model online. First, a weighted average strategy is adopted to update the reference jitter feature template that matches the current operating conditions using the measured jitter feature vectors.

[0054] Secondly, once a jitter event is confirmed as valid, a model update operation is performed. This invention preferably employs a weighted average strategy to update the reference jitter feature template. The operating conditions of the hoist, i.e., the electromechanical operating parameter vector, are then updated. Clustering is performed; for example, the K-Means algorithm is used offline to divide historical operating data into C clusters, with each cluster corresponding to a reference jitter feature template. During online operation, for a new confirmed sample, its electromechanical parameters are set as follows: The actual jitter feature vector is ,calculate The distance between the center of all C historical load cases is used to find the nearest matching load case. .

[0055] Subsequently, the reference jitter feature template for this matching condition is updated using the exponentially weighted moving average (EWMA). ,Right now .in, This is the learning rate or update weights, a small value between 0 and 1, preferably set to 0.1 in this invention. Furthermore, for the model parameters of the prediction time window, a forgetting factor can be used. The Recursive Least Squares (RLS) algorithm is used for updating.

[0056] Specifically, when the initial confidence level is higher than the decision threshold, it is confirmed as a valid limit switch state switch and the corresponding interlock control command is triggered. That is, after confirming a valid limit switch state switch, a jitter-free and logically deterministic switch state signal is generated. The switch state signal is sent to the hoist safety control logic unit, which coordinates the actions of cutting off the drive motor power and engaging the braking system according to the preset safety procedures and the predetermined sequence, so as to realize the reliable stopping of the hoist.

[0057] In practice, once the initial confidence level of the measured jitter feature vector exceeds the dynamically set decision threshold, the adaptive de-jitter algorithm module immediately sets the internal logic state of the limit switch to the triggered-on state and outputs a stable and glitch-free digital signal. This digital signal is hardwired to the input terminal of the hoist safety control logic unit, such as a safety PLC or safety relay module.

[0058] Upon receiving a confirmed signal, the safety control logic unit immediately initiates a preset interlocking control sequence. For example, a typical safety parking sequence setting is as follows: Valid trigger signals are received at all times; Within the specified time frame, the safety control logic unit sends a Safety Torque Off (STO) command to the driver to cut off the power output of the drive motor; Within a specified time frame, the safety control logic unit disconnects the power supply to the mechanical brake coil, causing the spring-loaded mechanical brake to reliably engage in a de-energized state. The time difference is set to ensure that the motor torque has completely dissipated before the mechanical brake applies braking force, effectively preventing mechanical resistance between the motor and the brake, and thus reducing mechanical impact on the hoist.

[0059] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. An interlocking control method for automatic hoists based on limit switches, characterized in that, include: The system synchronously collects high-frequency level sequences of the limit switches of the hoist and electromechanical operating parameters during the hoist's operation. Based on historical electromechanical operating parameters and confirmed valid limit switch level jitter data, a prediction model is established. This prediction model predicts the expected trigger time window of the limit switches that matches the current operating condition based on the currently collected electromechanical operating parameters, and retrieves or generates reference jitter feature templates representing typical level jitter characteristics under the current operating condition. Within the expected trigger time window, when a level jump of the limit switch is detected, the transient level feature at the start of the jump is extracted to form the measured jitter feature vector. Calculate the similarity between the measured jitter feature vector and the reference jitter feature template to obtain an initial confidence level representing the degree of matching between the actual jitter and the expected jitter; set a decision threshold based on the historical prediction error statistics or real-time uncertainty assessment of the prediction model. When the initial confidence level is higher than the decision threshold, it is confirmed as a valid limit switch state switch change and the corresponding interlock control command is triggered.

2. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, The establishment of the prediction model includes: the prediction model is a state-space model, which takes the electromechanical operating parameters of the hoist as input, and takes the start and end points of the expected trigger time window and the reference jitter feature template as model output; the initial structure and parameters of the model are determined offline by an identification method, and the parameters are recursively corrected using new samples during online operation.

3. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, The step of extracting the transient level features at the start of the transition to form a measured jitter feature vector includes: taking the first level transition as the starting point, extracting a very short transient level sequence, such as 5ms, as an analysis segment; extracting time-domain features from the analysis segment, including the transition slope, transient pulse width, and effective level time ratio; and combining these time-domain feature parameters to form the measured jitter feature vector.

4. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, The calculation of the similarity between the measured jitter feature vector and the reference jitter feature template to obtain an initial confidence level representing the degree of matching between the actual jitter and the expected jitter includes: treating the measured jitter feature vector and the reference jitter feature template as two points in a multi-dimensional feature space, and calculating the weighted Euclidean distance or Mahalanobis distance between the two points; converting the calculated distance value into a similarity value between 0 and 1 through a preset mapping function, and using it as the initial confidence level.

5. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, The method of statistical analysis of historical prediction errors or assessment of real-time uncertainty based on the prediction model and setting a decision threshold includes: tracking and statistically analyzing the deviation between the predicted value of the prediction model for the center point of the expected trigger time window and the actual confirmed trigger time point in real time, calculating the standard deviation in the most recent time period; normalizing the standard deviation by dividing it by a preset reference time value, multiplying it by a preset risk coefficient, and subtracting the multiplication result from a fixed basic threshold between 0 and 1. If the result is less than a preset minimum threshold, the minimum threshold is taken, and the result is the final decision threshold under the current working condition.

6. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, Also includes: The confirmed effective measured jitter feature vectors are used as new samples to update the prediction model online; specifically, a weighted average strategy is adopted to update the reference jitter feature template that matches the current operating conditions using the measured jitter feature vectors.

7. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, When the initial confidence level is higher than the decision threshold, it is confirmed as a valid limit switch state switch and the corresponding interlock control command is triggered. This includes: after confirming a valid limit switch state switch, generating a jitter-free and logically deterministic switch state signal; sending the switch state signal to the hoist safety control logic unit, which coordinates the actions of cutting off the drive motor power and engaging the braking system according to the preset safety procedures and a predetermined sequence, so as to achieve a reliable stop of the hoist.

8. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, The synchronous acquisition of the high-frequency level sequence of the hoist limit switch and the electromechanical operating parameters during the hoist operation includes: acquiring the binary high and low level states of the limit switch through a digital input channel to form a high-frequency level sequence; simultaneously, synchronously reading electromechanical operating parameters including one or more combinations of speed, position, drive motor current or torque through a communication bus or analog input channel; all channel acquisition operations are triggered by a unified hardware clock signal.

9. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, The prediction model based on historical electromechanical operating parameters and confirmed valid limit switch level jitter data includes: using an unsupervised learning algorithm to perform cluster analysis on a large dataset of electromechanical operating parameters stored in the historical database, dividing the hoist's operating state into several typical working conditions; for each working condition category, constructing and training a long short-term memory network, the input of which is the time series of electromechanical operating parameters of the hoist under a specific working condition, and the output is the remaining time from the current moment until the hoist arrives and triggers the limit switch; extracting the time-frequency feature vectors of all triggering events under the same working condition category in the historical data, performing weighted average calculation, obtaining a reference jitter feature template specific to that working condition, and storing it in a template library.

10. The automatic hoist interlocking control method based on limit switches according to claim 1, characterized in that, Also includes: The confirmed and valid measured jitter feature vectors are used as new samples to update the prediction model online. Specifically, the confirmed and valid measured jitter feature vectors, the electromechanical parameter sequence before triggering, and the actual triggering time difference are used together as a new high-quality sample pair. The new high-quality sample pair is stored in an online learning buffer. When the number of samples in the buffer reaches the preset batch size, a small batch of online training is performed on the corresponding prediction model to fine-tune the network weights, so as to achieve continuous optimization of the prediction model.

Images