PLC programmable cement pole intelligent automatic demolding control method

Abstract

The application provides a PLC programmable cement pole intelligent automatic demolding control method, which comprises collecting mold stress, vibration frequency, servo motor torque and pole surface temperature multi-source data, mapping the vibration frequency and the mold stress into high-dimensional features by relying on a temperature-driven thermal perception sparse coding dictionary, fusing the torque and the temperature to construct a real-time data frame; reconstructing the memory layout by predicting the state transition entropy, continuously storing the high-correlation data frame to form a main operation data area, establishing an independent emergency disposal data partition to store safety control parameters; correcting the state transition matrix in real time by means of the stress and the torque, performing predictive access on the pre-fetch data cluster, and solving the optimal control sequence; the PLC issues a demolding control instruction according to the optimal sequence, calculates a normalized torque margin to implement hierarchical control: when the margin is in the early warning interval, the emergency data retrieval priority is improved and a fine tuning instruction is output, and when the margin is out of limit, emergency braking is performed to ensure the safety of the demolding operation equipment and components.

Description

Technical Field

[0001] This application belongs to the field of cement poles, and in particular relates to a PLC programmable intelligent automatic demolding control method for cement poles. Background Technology

[0002] As a crucial component in the construction of power and communication infrastructure, the demolding process of cement poles directly impacts product quality and production efficiency. In actual demolding operations, the stress and separation process between the mold and the pole is complex, typically accompanied by drastic changes in mold stress, high-frequency vibration, servo motor torque fluctuations, and material surface temperature variations. Existing cement pole demolding systems largely rely on manual experience or conventional control equipment, making it difficult to fuse and analyze multi-source heterogeneous state data during demolding and to perceive it in real time. This results in the system's inability to accurately identify the demolding status. When demolding conditions fluctuate or uneven local stress occurs, insufficient control precision can easily cause scratches, damage, or internal micro-cracks on the cement pole surface. In severe cases, it can even cause irreversible deformation and damage to the demolding mold, affecting the intelligence level and overall yield of the cement pole production line.

[0003] With the continuous integration of Industrial Internet of Things (IIoT) and intelligent manufacturing technologies, the introduction of multi-sensor monitoring and intelligent control algorithms has become an important direction for optimizing the demolding process of cement poles. Demolding control not only requires the system to assess current operational risks in real time, but also to generate and issue reasonable operation sequences within a short control cycle. Existing control systems, when processing large amounts of state data, typically still employ traditional data storage and retrieval mechanisms, making it difficult to predictively optimize the layout of memory or cached data based on the transition patterns of demolding states. Consequently, in sudden dangerous situations such as motor torque instantaneously approaching safety limits, the retrieval of critical operational data and emergency commands is prone to delays. Lag in computation and data scheduling makes it difficult for actuators to trigger preventative fine-tuning responses or emergency braking in a timely manner, thereby increasing the risk of equipment overload and safety accidents. Therefore, there is an urgent need for a demolding control method that combines underlying cached data layout optimization with multi-source data intelligent predictive control to address the problems of insufficient real-time performance and delayed emergency response in demolding control systems under varying operating conditions. Summary of the Invention

[0004] According to a first aspect of the present invention, a PLC-programmable intelligent automatic demolding control method for cement poles is proposed to solve the problems of poor real-time performance and delayed emergency response of demolding control systems under varying working conditions, comprising the following steps: Acquire multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque, and pole surface temperature. Utilize a thermal sensing sparse coding dictionary indexed by pole surface temperature to convert the dimensionless vibration frequency and mold stress into high-dimensional feature vectors. Then, aggregate the high-dimensional feature vectors with the servo motor torque and pole surface temperature into a real-time data frame. The memory layout is reconstructed based on the predicted state transition entropy. Real-time data frames related to the subsequent highest probability state are arranged continuously in memory to obtain a prefetched data cluster as the main operation data area. An emergency response data partition is established to store the limit torque threshold, emergency braking command, safe load reduction mode parameters and their default priority weights. Based on the context state transition matrix that is calibrated in real time by the current mold stress and the current servo motor torque, the prefetched data cluster is accessed in advance. The risk assessment value and the optimal operation sequence are calculated by combining the candidate control action sequence. Based on the optimal operation sequence, control instructions for the demolding actuator are generated. The normalized torque margin between the current servo motor torque and the limit torque threshold is calculated. A graded response is executed. When the normalized torque margin is lower than the safety threshold but greater than 0, the priority of the emergency response data partition parameters is increased by using the normalized torque margin as an inverse factor, and a preventive fine-tuning control instruction is generated and issued for execution. When the normalized torque margin is not greater than 0, an emergency braking instruction is executed.

[0005] Optionally, the acquisition of multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque, and pole surface temperature, and the conversion of the dimensionless vibration frequency and mold stress into high-dimensional feature vectors using a thermal sensing sparse coding dictionary indexed by the pole surface temperature, includes: The surface temperature of the pole is divided into multiple preset temperature ranges, and each temperature range is mapped to a corresponding local thermal sensing sparse coding dictionary. Determine the target temperature range to which the currently acquired pole surface temperature belongs, and extract the target thermal sensing sparse coding dictionary corresponding to the target temperature range; The vibration frequency and mold stress are respectively dimensionless to form a two-dimensional initial feature vector. The target thermal sensing sparse coding dictionary is used to perform sparse coding operations on the two-dimensional initial feature vector, and the mapped high-dimensional feature vector is output.

[0006] Optionally, the aggregation of the high-dimensional feature vector with the servo motor torque and the pole surface temperature into a real-time data frame includes: Allocate a contiguous memory space of a preset byte length in the system heap memory as the data frame construction area; Write the high-dimensional feature vectors into the starting address segment of the data frame construction area in dimensional order; The values ​​of the servo motor torque and the pole surface temperature are converted into normalized floating-point numbers and appended to the end address of the high-dimensional feature vector in the data frame construction area. A timestamp and a checksum are added to the end of the data frame construction area, and the real-time data frame is generated by encapsulation.

[0007] Optionally, the step of performing predictive access to the prefetched data cluster based on the context state transition matrix calibrated in real time using the current mold stress and the current servo motor torque, and calculating the risk assessment value and optimal operation sequence in combination with the candidate control action sequence, includes: Input the current mold stress and the current servo motor torque into a preset joint probability distribution model, and calculate the working condition deviation weight vector for each candidate subsequent state corresponding to the current combination mode. The transition probability elements of the static baseline state transition matrix are updated by weighting the working condition deviation weight vector, and the transition probabilities of each row after weighting are normalized to obtain the context state transition matrix. Based on the context state transition matrix, and combined with the preset candidate control action sequence, the candidate data frame ranked first in the prefetched data cluster is deduced to predict the state trajectory of the next multiple steps. The distance of the state trajectory from the safe state envelope in the normalized feature space is calculated as the risk assessment value. The candidate control action sequence that minimizes the risk assessment value is selected as the optimal operation sequence.

[0008] Optionally, the calculation of the normalized torque margin between the current servo motor torque and the ultimate torque threshold includes: The actual torque value of the servo motor at the current moment is read through the device port, and the absolute value of the actual torque value is taken to obtain the actual torque amplitude, which is then stored in the register. Call the preset limit torque threshold in the emergency response data partition; The difference between the ultimate torque threshold and the actual torque amplitude is calculated as the absolute torque margin; Divide the absolute torque margin by the ultimate torque threshold to obtain the normalized torque margin between the current servo motor torque and the ultimate torque threshold.

[0009] Optionally, when the normalized torque margin is lower than the safety threshold but greater than 0, the priority of pre-fetching emergency response data partition parameters is increased by using the normalized torque margin as an inverse factor, and a preventive fine-tuning control command is generated and issued for execution, including: The currently acquired normalized torque margin is numerically compared with the preset safety threshold. When the normalized torque margin is lower than the safety threshold and greater than 0, the reciprocal of the normalized torque margin is calculated, and the reciprocal is used as an inverse adjustment factor after being subjected to amplitude limiting. The inverse adjustment factor is multiplied by the default priority weight of the safety load reduction mode parameter in the emergency response data partition to obtain the updated high priority weight. The data retrieval priority of the safety load reduction mode parameter is increased according to the updated high priority weight, so that the safety load reduction mode parameter is given priority to enter the data queue to be accessed by the control program. Generate preventative fine-tuning control commands to reduce the servo motor speed and decrease the demolding thrust, and issue them for execution.

[0010] Optionally, executing the emergency braking command when the normalized torque margin is not greater than 0 includes: When the normalized torque margin is determined to be less than or equal to 0, a highest priority blocking signal is sent to the main controller. Suspend all routine operation processes of the current demolding actuator and clear the execution queue of routine control commands; The emergency braking instruction set is extracted from the independent emergency response data partition in memory and written into the underlying hardware control register of the demolding actuator to trigger the emergency stop action.

[0011] According to a second aspect of the present invention, a PLC-programmable intelligent automatic demolding control system for cement poles is provided, comprising the following modules: The conversion module is used to acquire multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque and pole surface temperature. Using the thermal sensing sparse coding dictionary indexed by pole surface temperature, the dimensionless vibration frequency and mold stress are converted into high-dimensional feature vectors, and the high-dimensional feature vectors are aggregated with servo motor torque and pole surface temperature into real-time data frames. The calculation module is used to reconstruct the memory layout based on the predicted state transition entropy. It arranges the real-time data frames related to the subsequent highest probability state continuously in memory to obtain the prefetched data cluster as the main operation data area. It also establishes an emergency response data partition that stores the limit torque threshold, emergency braking command, safe load reduction mode parameters and their default priority weights. Based on the context state transition matrix that is calibrated in real time by the current mold stress and the current servo motor torque, it performs predictive access to the prefetched data cluster and calculates the risk assessment value and the optimal operation sequence in combination with the candidate control action sequence. The execution module is used to generate control instructions for the demolding actuator based on the optimal operation sequence, calculate the normalized torque margin between the current servo motor torque and the limit torque threshold, and execute a graded response. When the normalized torque margin is lower than the safety threshold but greater than 0, the priority of the emergency response data partition parameters is increased by using the normalized torque margin as an inverse factor, and a preventive fine-tuning control instruction is generated and issued for execution. When the normalized torque margin is not greater than 0, an emergency braking instruction is executed.

[0012] Preferably, the acquisition of multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque, and pole surface temperature, utilizes a thermal sensing sparse coding dictionary indexed by the pole surface temperature to convert the dimensionless vibration frequency and mold stress into high-dimensional feature vectors, including: The surface temperature of the pole is divided into multiple preset temperature ranges, and each temperature range is mapped to a corresponding local thermal sensing sparse coding dictionary. Determine the target temperature range to which the currently acquired pole surface temperature belongs, and extract the target thermal sensing sparse coding dictionary corresponding to the target temperature range; The vibration frequency and mold stress are respectively dimensionless to form a two-dimensional initial feature vector. The target thermal sensing sparse coding dictionary is used to perform sparse coding operations on the two-dimensional initial feature vector, and the mapped high-dimensional feature vector is output.

[0013] Preferably, the step of aggregating the high-dimensional feature vector with the servo motor torque and the pole surface temperature into a real-time data frame includes: Allocate a contiguous memory space of a preset byte length in the system heap memory as the data frame construction area; Write the high-dimensional feature vectors into the starting address segment of the data frame construction area in dimensional order; The values ​​of the servo motor torque and the pole surface temperature are converted into normalized floating-point numbers and appended to the end address of the high-dimensional feature vector in the data frame construction area. A timestamp and a checksum are added to the end of the data frame construction area, and the real-time data frame is generated by encapsulation.

[0014] Preferably, the step of performing predictive access to the prefetched data cluster based on the context state transition matrix calibrated in real time using the current mold stress and the current servo motor torque, and calculating the risk assessment value and optimal operation sequence in combination with the candidate control action sequence, includes: Input the current mold stress and the current servo motor torque into a preset joint probability distribution model, and calculate the working condition deviation weight vector for each candidate subsequent state corresponding to the current combination mode. The transition probability elements of the static baseline state transition matrix are updated by weighting the working condition deviation weight vector, and the transition probabilities of each row after weighting are normalized to obtain the context state transition matrix. Based on the context state transition matrix, and combined with the preset candidate control action sequence, the candidate data frame ranked first in the prefetched data cluster is deduced to predict the state trajectory of the next multiple steps. The distance of the state trajectory from the safe state envelope in the normalized feature space is calculated as the risk assessment value. The candidate control action sequence that minimizes the risk assessment value is selected as the optimal operation sequence.

[0015] Preferably, the calculation of the normalized torque margin between the current servo motor torque and the ultimate torque threshold includes: The actual torque value of the servo motor at the current moment is read through the device port, and the absolute value of the actual torque value is taken to obtain the actual torque amplitude, which is then stored in the register. Call the preset limit torque threshold in the emergency response data partition; The difference between the ultimate torque threshold and the actual torque amplitude is calculated as the absolute torque margin; Divide the absolute torque margin by the ultimate torque threshold to obtain the normalized torque margin between the current servo motor torque and the ultimate torque threshold.

[0016] Preferably, when the normalized torque margin is lower than the safety threshold but greater than 0, the step of increasing the pre-fetching priority of emergency response data partition parameters and generating a preventative fine-tuning control command for execution, using the normalized torque margin as an inverse factor, includes: The currently acquired normalized torque margin is numerically compared with the preset safety threshold. When the normalized torque margin is lower than the safety threshold and greater than 0, the reciprocal of the normalized torque margin is calculated, and the reciprocal is used as an inverse adjustment factor after being subjected to amplitude limiting. The inverse adjustment factor is multiplied by the default priority weight of the safety load reduction mode parameter in the emergency response data partition to obtain the updated high priority weight. The data retrieval priority of the safety load reduction mode parameter is increased according to the updated high priority weight, so that the safety load reduction mode parameter is given priority to enter the data queue to be accessed by the control program. Generate preventative fine-tuning control commands to reduce the servo motor speed and decrease the demolding thrust, and issue them for execution.

[0017] Preferably, the step of executing the emergency braking command when the normalized torque margin is not greater than 0 includes: When the normalized torque margin is determined to be less than or equal to 0, a highest priority blocking signal is sent to the main controller. Suspend all routine operation processes of the current demolding actuator and clear the execution queue of routine control commands; The emergency braking instruction set is extracted from the independent emergency response data partition in memory and written into the underlying hardware control register of the demolding actuator to trigger the emergency stop action.

[0018] This invention enhances the system's real-time perception of demolding status by collecting mold stress, vibration frequency, servo motor torque, and pole surface temperature, and utilizing a thermal sensing sparse coding dictionary to perform high-dimensional feature transformation and data aggregation. Based on predicted state transition entropy, the memory layout is reconstructed, arranging high-probability related data frames continuously to form pre-fetched data clusters, and setting up independent emergency response data partitions, reducing critical data reading latency and improving the control system's data processing efficiency and response speed. Simultaneously, predictive data access is performed using a state transition matrix with real-time calibration of transition probabilities, quickly generating optimal operation sequences, and executing graded responses in conjunction with normalized torque margins. In high-risk operating conditions, the priority of emergency data access can be promptly increased and preventative fine-tuning control implemented, immediately triggering emergency braking when torque exceeds limits, thereby reducing the risk of mechanical equipment damage and pole breakage, and improving the safety protection capabilities and overall production quality of cement pole demolding. Attached Figure Description

[0019] Figure 1 A flowchart of the first embodiment; Figure 2 This is a schematic diagram of the analysis of 256-dimensional feature vectors; Figure 3 A diagram illustrating the triggering of the fine-tuning load reduction safety action. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In the first embodiment, the present invention proposes a PLC-programmable intelligent automatic demolding control method for cement poles, such as... Figure 1 It includes the following steps: S1. Acquire multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque, and pole surface temperature. Using a thermal sensing sparse coding dictionary indexed by pole surface temperature, convert the dimensionless vibration frequency and mold stress into high-dimensional feature vectors, and aggregate the high-dimensional feature vectors with servo motor torque and pole surface temperature into real-time data frames.

[0022] Mold stress signals are acquired using fiber Bragg grating sensors deployed on the demolding machine, vibration signals are acquired using piezoelectric accelerometers, and vibration frequency characteristics are extracted from these signals. Torque signals are acquired using a torque sensor built into a servo motor, and the surface temperature of the pole is acquired using an infrared thermal imager. The multi-source analog signals are converted into one-dimensional time-series digital signals using an acquisition module or analog-to-digital converter. These one-dimensional time-series digital signals are then transmitted to an industrial computer or edge controller using the industrial Ethernet bus protocol and the SOEM open-source master station library based on the C language. Within each control cycle, the one-dimensional time-series digital signals are statistically extracted according to a preset sliding time window. Specifically, the dominant frequency, peak frequency, or spectral energy center is extracted from the vibration signal as the current vibration frequency scalar, and the window mean, peak value, or root mean square value is extracted from the mold stress signal as the current mold stress scalar. Both the current vibration frequency scalar and the current mold stress scalar are then dimensionless according to their respective calibrated ranges.

[0023] A thermal sensing sparse coding dictionary based on a hash table data structure is pre-constructed in the industrial control computer's memory. Discretized temperature ranges are used as hash keys and mapped to a pre-trained feature dictionary matrix. A hash lookup is performed based on the currently acquired pole surface temperature value to extract a matching feature dictionary matrix. An orthogonal matching pursuit algorithm function is called, using the extracted feature dictionary matrix as a base. The dimensionless current mold stress scalar and the current vibration frequency scalar are combined to form a two-dimensional initial feature vector. Sparse coding operations are then performed on this two-dimensional initial feature vector to obtain a high-dimensional feature vector with sparse characteristics. The timestamp, high-dimensional feature vector, servo motor torque, and pole surface temperature are memory-aligned to integer multiples of 64 bytes for cache line size and packaged into a continuous real-time data frame object.

[0024] In an optional embodiment, the acquisition of multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque, and pole surface temperature, and the conversion of the dimensionless vibration frequency and mold stress into high-dimensional feature vectors using a thermal sensing sparse coding dictionary indexed by the pole surface temperature, includes: The surface temperature of the pole is divided into multiple preset temperature ranges, and each temperature range is mapped to a corresponding local thermal sensing sparse coding dictionary. Determine the target temperature range to which the currently acquired pole surface temperature belongs, and extract the target thermal sensing sparse coding dictionary corresponding to the target temperature range; The vibration frequency and mold stress are respectively dimensionless to form a two-dimensional initial feature vector. The target thermal sensing sparse coding dictionary is used to perform sparse coding operations on the two-dimensional initial feature vector, and the mapped high-dimensional feature vector is output.

[0025] In the feature extraction and dictionary mapping stage, a correspondence is established between the pole surface temperature and the sparse dictionary. Common operating temperature ranges on the pole surface, such as 0℃ to 100℃, are divided into continuous preset temperature intervals [0,20), [20,40), and [40,60), with a step size of 20℃. During the offline training stage, historical mold state data for each temperature interval are jointly optimized to generate a 2×256-dimensional local thermal sensing sparse coding dictionary matrix, which is then mapped to the static lookup table for each interval. During real-time demolding operations, based on the identified pole surface temperature, for example, 35.5℃, the target temperature interval [20,40) is determined, and the 2×256-dimensional target thermal sensing sparse coding dictionary mapped to this interval is extracted from the lookup table in memory and loaded into the cache. The control system extracts the main frequency of the synchronously acquired high-frequency vibration signal within the current control window, ranging from 0 to 500 Hz. For example, the current main frequency is 45.2 Hz. It also extracts the root mean square value or mean value of the low-frequency mold stress signal within the current control window, ranging from 0 to 100 MPa. For example, the current stress characteristic value is 28.4 MPa.

[0026] The control system uses the upper limit of vibration frequency calibration and the upper limit of mold stress calibration as normalization benchmarks, mapping 45.2Hz to 0.0904 and 28.4MPa to 0.284, respectively. These are then concatenated to form a 1×2 dimensional initial feature vector. To identify the thermal fusion effect, the extracted target thermal sensing sparse coding dictionary is used, and an orthogonal matching pursuit algorithm is employed to perform sparse coding operations on the 2D initial feature vector. The sparsity limit K is set to not exceed the upper limit of the rank of the 2D observation space; in this example, K is set to 2, or it can be adaptively selected as a positive integer not greater than 2 based on the effective rank of the dictionary. After inner product matching and residual iterative updates, the low-dimensional signal is converted into a high-dimensional feature vector output consisting of a linear combination of a small number of effective atoms from the 256 dictionary atoms, achieving decoupling and high-dimensional representation of the joint vibration-stress features under different thermal environments. The distribution law of the 256-dimensional feature weights is as follows: Figure 2 As shown.

[0027] In an optional embodiment, the aggregation of the high-dimensional feature vector with the servo motor torque and the pole surface temperature into a real-time data frame includes: Allocate a contiguous memory space of a preset byte length in the system heap memory as the data frame construction area; Write the high-dimensional feature vectors into the starting address segment of the data frame construction area in dimensional order; The values ​​of the servo motor torque and the pole surface temperature are converted into normalized floating-point numbers and appended to the end address of the high-dimensional feature vector in the data frame construction area. A timestamp and a checksum are added to the end of the data frame construction area, and the real-time data frame is generated by encapsulation.

[0028] During the system memory allocation phase, a fixed-length, pre-defined 1088-byte contiguous memory space is allocated in the system heap memory as the data frame construction area. This 1088-byte length is an integer multiple of the 64-byte cache line size, ensuring memory address alignment and data read / write efficiency. The 256-dimensional high-dimensional feature vector output through sparse coding operations, if each element is a 32-bit single-precision floating-point number, occupies a total of 1024 bytes. This vector is written to the starting address range 0x0000 to 0x03FF of the data frame construction area, in the order of feature dimensions. The current servo motor torque (e.g., 150.5 N·m) and pole surface temperature (e.g., 45.2℃) are collected. The raw values ​​of these two parameters are converted into 32-bit normalized floating-point numbers conforming to the IEEE 754 standard and written to the last contiguous address range after the high-dimensional feature vector in the construction area, i.e., 0x0400 to 0x0407.

[0029] To ensure data transmission security and time-trackability, a 64-bit timestamp is appended to the end of the data frame construction area, in address segments 0x0408 to 0x040F. For example, this timestamp records the current system time as 1683012345678. A 16-bit Cyclic Redundancy Check (CRC) code (e.g., 0x4B3A) is calculated for all preceding valid fields, and this CRC-16 is written to address segments 0x0410 to 0x0411. Address segments 0x0412 to 0x043F are used as alignment reserve fields, filled with preset fixed values ​​or zero values. Therefore, the effective payload length of the real-time data frame is 1042 bytes, the actual allocated length is 1088 bytes, and the remaining 46 bytes are used for 64-byte multiples of alignment padding. In this way, the encapsulated real-time data frame has a fixed-length memory structure, supporting memory-based prefetching and access.

[0030] S2, based on the predicted state transition entropy, reconstructs the memory layout, arranges the real-time data frames related to the subsequent highest probability state continuously in memory, obtains the prefetched data cluster as the main operation data area, and establishes an emergency response data partition that stores the limit torque threshold, emergency braking command, safe load reduction mode parameters and their default priority weights. Based on the context state transition matrix that is calibrated in real time by the current mold stress and the current servo motor torque, the prefetched data cluster is accessed in advance, and the risk assessment value and the optimal operation sequence are calculated by combining the candidate control action sequence.

[0031] In a demodulation control system, a discrete-time Markov decision process model is constructed to calculate the information entropy of the transition probability distribution from the current system state to all possible subsequent states, and to measure the uncertainty of state transitions. The information entropy is divided by the maximum entropy to obtain the normalized predicted state transition entropy, where the maximum entropy is the logarithm of the number of candidate subsequent states. Based on the normalized predicted state transition entropy, the number of candidate states and the width of the prefetching window are determined. When the normalized predicted state transition entropy is low, it indicates that subsequent states are concentrated, and data frames are organized around the subsequent state corresponding to the highest state transition probability. When the normalized predicted state transition entropy is high, it indicates that subsequent states are dispersed, and data frames corresponding to multiple candidate subsequent states with high probability ranking are included in the prefetching data cluster and arranged continuously from high to low state transition probability. The mmap memory mapping system call interface is invoked to reconstruct memory. A contiguous page-aligned memory space is pre-allocated. Based on the prefetch window determined by the predicted state transition entropy and the main candidate state corresponding to the highest state transition probability, the historical and real-time data frames corresponding to the state with the high probability of occurrence in the next stage are migrated and stored sequentially in the contiguous memory space to form a prefetch data cluster. This data cluster is designated as the main operation data area.

[0032] In a separate emergency response data partition, fixed limit torque threshold constants, emergency braking command sets, safe load reduction mode parameters, and default priority weights of the safe load reduction mode parameters are written, enabling subsequent preventative fine-tuning and emergency braking to read the corresponding parameters from this partition. A basic Markov state transition matrix is ​​established. Within each control cycle, the normalized features of the current mold stress and the current servo motor torque are extracted, and the working condition deviation weight vectors for each candidate subsequent state corresponding to the current combination mode are calculated. These working condition deviation weight vectors are applied to the basic state transition matrix column-wise or by the direction of the subsequent state, and the transition probabilities of each row after weighted update are normalized to ensure that the sum of the transition probabilities of each row is 1, resulting in a real-time calibrated context state transition matrix. Based on the high-probability state memory addresses pointed to by the state transition matrix, data clusters in the main operation data area are pre-placed into the working set that the processor can quickly access through sequential prefetching, page prefetching hints, or reordering of the data queue to be accessed by the control program. The prefetching operation is used to improve the hit probability of subsequent access and reduce the average access latency, without limiting the data to be forcibly locked in the processor's L1 cache. Based on the extracted data, candidate control action sequences in a finite time domain are set. For each candidate control action sequence, future multi-step state deduction is performed one by one. The out-of-bounds penalty value of each predicted state deviating from the safe state envelope is calculated cumulatively to obtain the corresponding risk assessment value. The candidate control action sequence with the smallest cumulative risk assessment value is selected as the optimal operation sequence.

[0033] In an optional embodiment, the step of performing predictive access to prefetched data clusters based on a context state transition matrix calibrated in real time using the current mold stress and current servo motor torque, and calculating risk assessment values ​​and optimal operation sequences in conjunction with candidate control action sequences, includes: Input the current mold stress and the current servo motor torque into a preset joint probability distribution model, and calculate the working condition deviation weight vector for each candidate subsequent state corresponding to the current combination mode. The transition probability elements of the static baseline state transition matrix are updated by weighting the working condition deviation weight vector, and the transition probabilities of each row after weighting are normalized to obtain the context state transition matrix. Based on the context state transition matrix, and combined with the preset candidate control action sequence, the candidate data frame ranked first in the prefetched data cluster is deduced to predict the state trajectory of the next multiple steps. The distance of the state trajectory from the safe state envelope in the normalized feature space is calculated as the risk assessment value. The candidate control action sequence that minimizes the risk assessment value is selected as the optimal operation sequence.

[0034] The preset joint probability distribution model adopts a two-dimensional Gaussian mixture model, inputting the real-time acquired current mold stress (e.g., 35.5 MPa) and servo motor torque (e.g., 180.2 N·m) as two-dimensional vectors. By calculating the Mahalanobis distance between the input vector and the centers of various calibration working conditions, the model outputs a working condition deviation weight vector corresponding to each candidate subsequent state of the current combination mode. For example, different weights are assigned to candidate states such as normal release, slight jamming, and severe jamming, rather than simply outputting a single scalar coefficient. A 100×100 static baseline state transition matrix is ​​read from memory. The extracted working condition deviation weight vector is used to weight the elements representing the working condition transition probabilities in the matrix, and normalization is performed after weighting each row to ensure the sum of probabilities in each row is 1, generating a context state transition matrix adapted to the current demolding state. After real-time calibration of the transition probability matrix, a preset candidate control action sequence is loaded onto the top-ranked candidate data frame in the pre-fetched data cluster. This sequence may include five discrete action sets: demolding speed +10%, demolding speed -10%, and thrust -5%. For each candidate action sequence, a Markov decision process is used to predict the future time domain, for example, with a prediction step size of 5 and a prediction interval of 100ms per step, resulting in a multi-step state evolution trajectory. In a normalized feature space containing stress and torque, the sum or sum of squares of the positive deviations of the discrete state nodes of each predicted trajectory from the preset safe state envelope is calculated as the risk assessment value. When a state node is inside the safe state envelope, the corresponding deviation distance is recorded as 0; when a state node crosses the safe state envelope, the deviation distance increases with the degree of crossover, and the risk assessment value also increases accordingly. For example, the risk assessment value for trajectory A is 0.045, and for trajectory B it is 0.082, indicating that trajectory A is closer to the inside of the safe state envelope than trajectory B. The control system compares the simulation results of each action, selects the candidate control action sequence that minimizes the global risk assessment value, and outputs this sequence as the optimal operation sequence for the current control cycle.

[0035] S3 generates control instructions for the demolding actuator based on the optimal operation sequence, calculates the normalized torque margin between the current servo motor torque and the limit torque threshold, and executes a graded response. When the normalized torque margin is lower than the safety threshold but greater than 0, the priority of the emergency response data partition parameter pre-fetching is increased by using the normalized torque margin as an inverse factor, and a preventive fine-tuning control instruction is generated and issued for execution. When the normalized torque margin is not greater than 0, an emergency braking instruction is executed.

[0036] The digital control step size in the optimal operation sequence is converted into position and speed mode pulse commands recognizable by the servo driver. The current servo motor torque is read in real time, and the absolute value of the current servo motor torque is taken to obtain the current torque amplitude. The difference between the limit torque threshold read in the emergency response data partition and the current torque amplitude is calculated, and this difference is divided by the limit torque threshold to obtain the floating-point normalized torque margin. Multi-branch conditional judgment logic statements are written to execute hierarchical responses. When the normalized torque margin is less than the system-set safety threshold of 0.2 but greater than 0, the reciprocal of the normalized torque margin is calculated as an inverse proportional factor. After upper limit limiting of the inverse proportional factor, it is mapped to the priority of the control task or the priority of the data queue to be accessed for reading the emergency response data partition. The real-time scheduling interface is called to increase the scheduling level of the corresponding control task, and the safety deload mode parameters are given priority to enter the data queue to be accessed in the control program through memory pre-read prompts or data pre-loading mechanisms during PLC runtime. The above process is used to reduce subsequent reading latency, without limiting the memory pages of the emergency response data partition to be forcibly locked in the processor's L1 cache. Based on the inverse ratio factor, the servo motor speed setpoint and demolding thrust setpoint are adjusted downwards. Preventative fine-tuning control commands are generated to reduce the servo motor's operating speed and demolding thrust, and then sent to the motor driver for execution. When the normalized torque margin is detected to be equal to or less than 0, a safety interrupt or highest-priority safety task is immediately triggered, bypassing the normal control queue. The emergency braking command set is quickly read from the independent emergency response data partition, and a highest-priority service subroutine is triggered via a hardware timer. This resets the servo motor's enable signal to zero and activates the mechanical brake solenoid valve output interface, achieving braking.

[0037] In an optional embodiment, calculating the normalized torque margin between the current servo motor torque and the ultimate torque threshold includes: The actual torque value of the servo motor at the current moment is read through the device port, and the absolute value of the actual torque value is taken to obtain the actual torque amplitude, which is then stored in the register. Call the preset limit torque threshold in the emergency response data partition; The difference between the ultimate torque threshold and the actual torque amplitude is calculated as the absolute torque margin; Divide the absolute torque margin by the ultimate torque threshold to obtain the normalized torque margin between the current servo motor torque and the ultimate torque threshold.

[0038] During the servo motor torque monitoring cycle, the control program reads the actual torque value fed back by the servo driver at the current moment through the device port via the industrial field communication bus with a communication delay of 1 millisecond. Assuming the read actual servo motor torque is 245.5 N·m, the absolute value of this actual torque value is taken to obtain the actual torque amplitude of 245.5 N·m, and this actual torque amplitude is stored in the processor's internal data register, such as register R1. The pre-calibrated limit torque threshold in the emergency response data partition is then called. This threshold is set within the range of 120% to 150% of the motor's rated load torque; in this example, it is set to 300.0 N·m, and this threshold is loaded into the processor's register R2. The absolute torque margin is calculated by subtracting the actual torque amplitude of 245.5 N·m in register R1 from the limit torque threshold of 300.0 N·m in register R2; in this example, it is 54.5 N·m. Dividing the calculated absolute torque margin of 54.5 N·m by the ultimate torque threshold of 300.0 N·m yields the normalized torque margin, with an example result of 0.1817. This normalized index masks the absolute magnitude differences between different mold batches or different motor models, converting the torque safety margin into a dimensionless continuous index. This index is positive when the actual torque amplitude is below the ultimate torque threshold, and zero or negative when the actual torque amplitude reaches or exceeds the ultimate torque threshold.

[0039] In an optional embodiment, the step of increasing the pre-fetching priority of emergency response data partition parameters and generating a preventative fine-tuning control command for execution when the normalized torque margin is lower than the safety threshold but greater than 0 includes: The currently acquired normalized torque margin is numerically compared with the preset safety threshold. When the normalized torque margin is lower than the safety threshold and greater than 0, the reciprocal of the normalized torque margin is calculated, and the reciprocal is used as an inverse adjustment factor after being subjected to amplitude limiting. The inverse adjustment factor is multiplied by the default priority weight of the safety load reduction mode parameter in the emergency response data partition to obtain the updated high priority weight. The data retrieval priority of the safety load reduction mode parameter is increased according to the updated high priority weight, so that the safety load reduction mode parameter is given priority to enter the data queue to be accessed by the control program. Generate preventative fine-tuning control commands to reduce the servo motor speed and decrease the demolding thrust, and issue them for execution.

[0040] In the logic judgment stage of the control cycle, the normalized torque margin calculated in real time, for example 0.15, is compared with a preset safety threshold using floating-point numbers. This preset safety threshold is calibrated based on the equipment's fatigue strength and material elasticity level, ranging from 0.20 to 0.30; an example safety threshold is set to 0.25. When it is determined that the current normalized torque margin of 0.15 is lower than the safety threshold of 0.25 and simultaneously greater than 0, the reciprocal of the current normalized torque margin is calculated, i.e., 1 / 0.15 ≈ 6.67. This reciprocal is then limited by a preset upper limit to prevent abnormal amplification of the priority weight due to excessively small normalized torque margins. The safety load reduction mode parameter, located in the independent memory emergency response data partition, is invoked. The default priority weight pre-configured for this parameter is read, for example, the base weight is set to 10. This weight is multiplied by the inverse adjustment factor 6.67 to calculate the updated priority weight of 66.7.

[0041] Based on the updated priority weights, key parameter data blocks related to demolding, load reduction, and speed reduction are inserted into a high-priority data queue. These data blocks are prepared in advance through sequential pre-reading, memory page prefetching, or PLC runtime pre-loading mechanisms. These mechanisms increase the probability of timely reading of parameters in the safe load reduction mode. They are implemented through software scheduling and data access sequence optimization, without requiring the application to directly control the hardware cache controller or forcibly load data into the L1-level data cache. Based on the invoked load reduction parameters, control instructions are regenerated to reduce the target servo motor speed by 15%, for example, smoothly reducing it from the rated 150 rpm to 127.5 rpm, and to reduce the hydraulic cylinder demolding thrust by 10%, for example, limiting it from the upper limit of 50 kN to 45 kN as a preventative fine-tuning control instruction. This instruction is sent to the actuator via the fieldbus, completing the load reduction and force unloading action before mold jamming occurs. The speed and thrust fine-tuning load reduction change curves under the time delay window are shown below. Figure 3 As shown.

[0042] In an optional embodiment, executing the emergency braking command when the normalized torque margin is not greater than 0 includes: When the normalized torque margin is determined to be less than or equal to 0, a highest priority blocking signal is sent to the main controller. Suspend all routine operation processes of the current demolding actuator and clear the execution queue of routine control commands; The emergency braking instruction set is extracted from the independent emergency response data partition in memory and written into the underlying hardware control register of the demolding actuator to trigger the emergency stop action.

[0043] In the hardware assurance loop of demolding control, the status monitoring module performs safety boundary verification on the generated normalized torque margin. When the normalized torque margin is determined to be less than or equal to 0, for example, the latest calculation result is -0.03, it means that the actual servo motor torque amplitude has reached or exceeded the limit threshold that the equipment can withstand. Under this condition, the main controller triggers a safety interrupt or the highest priority safety task, sending a blocking signal with the highest priority to the control logic. With the blocking signal taking effect, the processor core executes a suspend instruction, suspending all normal operation processes of the current demolding actuator and freezing the motion state through a field context snapshot. The control task queue is cleared by writing a clear flag or resetting it, clearing the backlog of pending control instructions in the normal control instruction queue to prevent equipment damage caused by delayed instructions. The safety task accesses the address-isolated independent emergency response data partition according to the preset address index, extracts the emergency braking instruction set, including zero torque pulse setpoint, maximum energy consumption braking parameters, and hydraulic lock-up brake enable. This emergency instruction set is written to the hardware control register of the demolding actuator via the memory access channel. After the state of the identification register flips, the underlying logic of the actuator cuts off the power source output within a time window of no more than 5 milliseconds and activates the holding brake and emergency stop actions to ensure that the demolding machine can be stopped and the system safety is guaranteed.

[0044] In the second embodiment, the present invention also proposes a PLC programmable intelligent automatic demolding control system for cement poles, comprising the following modules: The conversion module is used to acquire multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque and pole surface temperature. Using the thermal sensing sparse coding dictionary indexed by pole surface temperature, the dimensionless vibration frequency and mold stress are converted into high-dimensional feature vectors, and the high-dimensional feature vectors are aggregated with servo motor torque and pole surface temperature into real-time data frames. The calculation module is used to reconstruct the memory layout based on the predicted state transition entropy. It arranges the real-time data frames related to the subsequent highest probability state continuously in memory to obtain the prefetched data cluster as the main operation data area. It also establishes an emergency response data partition that stores the limit torque threshold, emergency braking command, safe load reduction mode parameters and their default priority weights. Based on the context state transition matrix that is calibrated in real time by the current mold stress and the current servo motor torque, it performs predictive access to the prefetched data cluster and calculates the risk assessment value and the optimal operation sequence in combination with the candidate control action sequence. The execution module is used to generate control instructions for the demolding actuator based on the optimal operation sequence, calculate the normalized torque margin between the current servo motor torque and the limit torque threshold, and execute a graded response. When the normalized torque margin is lower than the safety threshold but greater than 0, the priority of the emergency response data partition parameters is increased by using the normalized torque margin as an inverse factor, and a preventive fine-tuning control instruction is generated and issued for execution. When the normalized torque margin is not greater than 0, an emergency braking instruction is executed.

[0045] The above description represents the preferred embodiments of the present invention. It should be noted that, for those skilled in the art, various improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A PLC-programmable intelligent automatic demolding control method for cement poles, characterized in that, Includes the following steps: Acquire multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque, and pole surface temperature. Utilize a thermal sensing sparse coding dictionary indexed by pole surface temperature to convert the dimensionless vibration frequency and mold stress into high-dimensional feature vectors. Then, aggregate the high-dimensional feature vectors with the servo motor torque and pole surface temperature into a real-time data frame. The memory layout is reconstructed based on the predicted state transition entropy. Real-time data frames related to the subsequent highest probability state are arranged continuously in memory to obtain a prefetched data cluster as the main operation data area. An emergency response data partition is established to store the limit torque threshold, emergency braking command, safe load reduction mode parameters and their default priority weights. Based on the context state transition matrix that is calibrated in real time by the current mold stress and the current servo motor torque, the prefetched data cluster is accessed in advance. The risk assessment value and the optimal operation sequence are calculated by combining the candidate control action sequence. Based on the optimal operation sequence, control instructions for the demolding actuator are generated. The normalized torque margin between the current servo motor torque and the limit torque threshold is calculated. A graded response is executed. When the normalized torque margin is lower than the safety threshold but greater than 0, the priority of the emergency response data partition parameters is increased by using the normalized torque margin as an inverse factor, and a preventive fine-tuning control instruction is generated and issued for execution. When the normalized torque margin is not greater than 0, an emergency braking instruction is executed.

2. The method according to claim 1, characterized in that, The acquisition of multi-source state data during the demolding process includes mold stress, vibration frequency, servo motor torque, and pole surface temperature. Using a thermal sensing sparse coding dictionary indexed by the pole surface temperature, the dimensionless vibration frequency and mold stress are converted into high-dimensional feature vectors, including: The surface temperature of the pole is divided into multiple preset temperature ranges, and each temperature range is mapped to a corresponding local thermal sensing sparse coding dictionary. Determine the target temperature range to which the currently acquired pole surface temperature belongs, and extract the target thermal sensing sparse coding dictionary corresponding to the target temperature range; The vibration frequency and mold stress are respectively dimensionless to form a two-dimensional initial feature vector. The target thermal sensing sparse coding dictionary is used to perform sparse coding operations on the two-dimensional initial feature vector, and the mapped high-dimensional feature vector is output.

3. The method according to claim 2, characterized in that, The process of aggregating high-dimensional feature vectors with servo motor torque and pole surface temperature into a real-time data frame includes: Allocate a contiguous memory space of a preset byte length in the system heap memory as the data frame construction area; Write the high-dimensional feature vectors into the starting address segment of the data frame construction area in dimensional order; The values ​​of the servo motor torque and the pole surface temperature are converted into normalized floating-point numbers and appended to the end address of the high-dimensional feature vector in the data frame construction area. A timestamp and a checksum are added to the end of the data frame construction area, and the real-time data frame is generated by encapsulation.

4. The method according to claim 1, characterized in that, The context state transition matrix, based on the real-time calibration of the transition probability using the current mold stress and the current servo motor torque, performs predictive access to the pre-fetched data cluster and calculates the risk assessment value and optimal operation sequence in conjunction with candidate control action sequences, including: Input the current mold stress and the current servo motor torque into a preset joint probability distribution model, and calculate the working condition deviation weight vector for each candidate subsequent state corresponding to the current combination mode. The transition probability elements of the static baseline state transition matrix are updated by weighting the working condition deviation weight vector, and the transition probabilities of each row after weighting are normalized to obtain the context state transition matrix. Based on the context state transition matrix, and combined with the preset candidate control action sequence, the candidate data frame ranked first in the prefetched data cluster is deduced to predict the state trajectory of the next multiple steps. The distance of the state trajectory from the safe state envelope in the normalized feature space is calculated as the risk assessment value. The candidate control action sequence that minimizes the risk assessment value is selected as the optimal operation sequence.

5. The method according to claim 1 or 4, characterized in that, The calculation of the normalized torque margin between the current servo motor torque and the ultimate torque threshold includes: The actual torque value of the servo motor at the current moment is read through the device port, and the absolute value of the actual torque value is taken to obtain the actual torque amplitude, which is then stored in the register. Call the preset limit torque threshold in the emergency response data partition; The difference between the ultimate torque threshold and the actual torque amplitude is calculated as the absolute torque margin; Divide the absolute torque margin by the ultimate torque threshold to obtain the normalized torque margin between the current servo motor torque and the ultimate torque threshold.

6. The method according to claim 1, characterized in that, When the normalized torque margin is lower than the safety threshold but greater than 0, the priority of pre-fetching emergency response data partition parameters is increased by using the normalized torque margin as an inverse factor, and a preventive fine-tuning control command is generated and issued for execution, including: The currently acquired normalized torque margin is numerically compared with the preset safety threshold. When the normalized torque margin is lower than the safety threshold and greater than 0, the reciprocal of the normalized torque margin is calculated, and the reciprocal is used as an inverse adjustment factor after being subjected to amplitude limiting. The inverse adjustment factor is multiplied by the default priority weight of the safety load reduction mode parameter in the emergency response data partition to obtain the updated high priority weight. The data retrieval priority of the safety load reduction mode parameter is increased according to the updated high priority weight, so that the safety load reduction mode parameter is given priority to enter the data queue to be accessed by the control program. Generate preventative fine-tuning control commands to reduce the servo motor speed and decrease the demolding thrust, and issue them for execution.

7. The method according to claim 1, characterized in that, When the normalized torque margin is not greater than 0, the emergency braking command is executed, including: When the normalized torque margin is determined to be less than or equal to 0, a highest priority blocking signal is sent to the main controller. Suspend all routine operation processes of the current demolding actuator and clear the execution queue of routine control commands; The emergency braking instruction set is extracted from the independent emergency response data partition in memory and written into the underlying hardware control register of the demolding actuator to trigger the emergency stop action.

8. A PLC-programmable intelligent automatic demolding control system for cement poles, characterized in that, Includes the following modules: The conversion module is used to acquire multi-source state data during the demolding process, including mold stress, vibration frequency, servo motor torque and pole surface temperature. Using the thermal sensing sparse coding dictionary indexed by pole surface temperature, the dimensionless vibration frequency and mold stress are converted into high-dimensional feature vectors, and the high-dimensional feature vectors are aggregated with servo motor torque and pole surface temperature into real-time data frames. The calculation module is used to reconstruct the memory layout based on the predicted state transition entropy. It arranges the real-time data frames related to the subsequent highest probability state continuously in memory to obtain the prefetched data cluster as the main operation data area. It also establishes an emergency response data partition that stores the limit torque threshold, emergency braking command, safe load reduction mode parameters and their default priority weights. Based on the context state transition matrix that is calibrated in real time by the current mold stress and the current servo motor torque, it performs predictive access to the prefetched data cluster and calculates the risk assessment value and the optimal operation sequence in combination with the candidate control action sequence. The execution module is used to generate control instructions for the demolding actuator based on the optimal operation sequence, calculate the normalized torque margin between the current servo motor torque and the limit torque threshold, and execute a graded response. When the normalized torque margin is lower than the safety threshold but greater than 0, the priority of the emergency response data partition parameters is increased by using the normalized torque margin as an inverse factor, and a preventive fine-tuning control instruction is generated and issued for execution. When the normalized torque margin is not greater than 0, an emergency braking instruction is executed.

9. The system according to claim 8, characterized in that, The acquisition of multi-source state data during the demolding process includes mold stress, vibration frequency, servo motor torque, and pole surface temperature. Using a thermal sensing sparse coding dictionary indexed by the pole surface temperature, the dimensionless vibration frequency and mold stress are converted into high-dimensional feature vectors, including: The surface temperature of the pole is divided into multiple preset temperature ranges, and each temperature range is mapped to a corresponding local thermal sensing sparse coding dictionary. Determine the target temperature range to which the currently acquired pole surface temperature belongs, and extract the target thermal sensing sparse coding dictionary corresponding to the target temperature range; The vibration frequency and mold stress are respectively dimensionless to form a two-dimensional initial feature vector. The target thermal sensing sparse coding dictionary is used to perform sparse coding operations on the two-dimensional initial feature vector, and the mapped high-dimensional feature vector is output.

10. The system according to claim 8, characterized in that, The process of aggregating high-dimensional feature vectors with servo motor torque and pole surface temperature into a real-time data frame includes: Allocate a contiguous memory space of a preset byte length in the system heap memory as the data frame construction area; Write the high-dimensional feature vectors into the starting address segment of the data frame construction area in dimensional order; The values ​​of the servo motor torque and the pole surface temperature are converted into normalized floating-point numbers and appended to the end address of the high-dimensional feature vector in the data frame construction area. A timestamp and a checksum are added to the end of the data frame construction area, and the real-time data frame is generated by encapsulation.

Images