Multi-sensor self-adaptive control system and method of injection-stretch-blow molding machine

Abstract

The invention relates to the technical field of industrial control systems, in particular to a multi-sensor self-adaptive control system and method of an injection-stretch-blow molding machine. The intelligent diagnosis module performs rapid detection and fine working condition identification by adopting a layered architecture, and outputs a diagnosis identifier and confidence; the decision-making module fuses expert knowledge and an online learning result, and determines an optimization coping strategy in combination with confidence; dynamically loading a control parameter set according to a strategy; the adaptive control module calculates an original instruction based on the parameter set and the data; generating a final signal according to the confidence coefficient smoothing instruction; and the execution interface module sends the signal to an execution mechanism to realize self-adaptive control. The invention further discloses a corresponding method. According to the system and the method, accurate diagnosis, differential decision and accurate strategy execution can be realized, smooth transition of control is ensured, and the control performance, the product quality and the operation robustness are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial control systems, and in particular to a multi-sensor adaptive control system and method for an injection stretch blow molding machine. Background Art

[0002] In the automation control of complex industrial processes, the use of program control systems to achieve basic process automation and the introduction of adaptive control technology to cope with process parameter changes and external disturbances have become a common means to improve production stability and efficiency. The application of multi-sensor technology also makes it possible to obtain richer process information.

[0003] However, despite the progress made in existing technologies, there is still a prominent core technology bottleneck when dealing with highly complex, nonlinear, time-varying industrial processes that are susceptible to multiple unexpected disturbances. For example, this bottleneck is particularly prominent in precision molding processes such as injection stretch blow molding, which require extremely high product accuracy and consistency. Specifically, current adaptive control systems, even if they utilize multi-sensor information, generally lack the ability to deeply understand and distinguish different types of operating conditions. It is difficult for the system to intelligently identify the root cause of performance deviations from complex data.

[0004] It is precisely because of the lack of this differentiated cognitive ability based on accurate working condition diagnosis that the existing adaptive control adjustment strategies often appear too stereotyped. They find it difficult to dynamically adjust their core control logic, optimization goals, and adaptation mechanisms in a targeted manner based on the specific root causes of problems diagnosed in real time. The blindness and roughness of this adaptability greatly limit the optimal performance and long-term robustness that the control system can achieve in a real and changing industrial environment. Therefore, developing a control method that can accurately diagnose working conditions and perform intelligent, differentiated adaptive adjustments based on them has become a key issue that needs to be urgently addressed in the field of advanced industrial control.

[0005] Therefore, a multi-sensor adaptive control system and method for injection stretch blow molding machine is proposed. Summary of the invention

[0006] The object of the present invention is to provide a multi-sensor adaptive control system and method for an injection stretch blow molding machine, which can significantly improve control performance, product quality and operation robustness by realizing accurate diagnosis, differential decision-making and accurate strategy execution, and ensuring a smooth transition of control. First, multi-modal data is collected through a sensor module; then, a smart diagnosis module uses a hierarchical architecture for rapid detection and fine working condition identification, and outputs a diagnosis flag and confidence level; next, a decision-making module fuses expert knowledge and online learning results, and determines an optimized coping strategy in combination with the confidence level; the control parameter set is dynamically loaded according to the strategy; subsequently, an adaptive control module calculates the original instruction based on the parameter set and data; the final signal is generated by smoothing the instruction according to the confidence level; finally, the signal is sent to the actuator through an execution interface module.

[0007] To achieve the above object, the present invention provides the following technical solutions: A multi-sensor adaptive control system for an injection stretch blow molding machine, comprising: A sensor module, which collects multi-modal process data and performs preprocessing; A data management module, which stores and retrieves the process data, models, strategies, knowledge bases and learning results required for operation; An interaction module, which monitors the system status and receives parameter inputs; A smart diagnosis module, which analyzes the data from the sensor module, uses the machine learning model stored in the data management module to identify the operating conditions, and generates a diagnosis flag and a diagnosis confidence level; A decision-making module, which receives the diagnosis flag, combines the expert knowledge and online learning results stored in the data management module, and determines an optimized adaptive coping strategy; according to the determined coping strategy, the relevant control parameters for subsequent calculations are dynamically adjusted; An adaptive control module, which calculates the original control instruction based on the adjusted relevant control parameters and the data from the sensor module; based on the diagnosis confidence level and the original control instruction, a final control signal that is smoothed to ensure a stable transition is generated; An execution interface module, which receives the final control signal and sends it to the corresponding actuator of the device.

[0008] Preferably, the smart diagnosis module adopts a hierarchical diagnosis structure, including: a first diagnosis layer for rapid anomaly detection and a second diagnosis layer for fine working condition classification; the smart diagnosis module integrates an interpretable artificial intelligence analysis function, which is used to provide the input features and the explanation information of the sensor source that have a significant impact on the results after diagnosis.

[0009] Preferably, the determination of the optimized adaptive coping strategy includes: Retrieve a benchmark response strategy that matches the current diagnostic condition from the expert knowledge base stored in the data management module based on the diagnostic identifier generated by the intelligent diagnostic module; Input the diagnostic identifier as status information into the online learning module stored in the data management module, and obtain a recommended response strategy for the current diagnostic condition learned by the online learning module based on historical operation data and long-term performance feedback; Integrate the benchmark response strategy and the recommended response strategy according to a preset fusion algorithm, and consider the confidence score of the diagnostic identifier to generate an optimized adaptive response strategy for final output.

[0010] Preferably, the preset fusion algorithm adopts a dynamic weighted fusion method, and the determination logic of its weight coefficient includes: initially determining a benchmark weight value according to the current diagnostic confidence score, and generating an adjustment factor by using the real-time analysis result of historical performance data, and the adjustment factor is used to adjust the benchmark weight value.

[0011] Preferably, the dynamic adjustment implemented by the decision-making module includes: Multiple groups of different control parameter configuration sets corresponding to different response strategy identifiers are pre-stored in the data management module; the decision-making module retrieves and loads the corresponding complete parameter configuration set from the data management module according to the received current response strategy identifier.

[0012] Preferably, the adaptive control module adopts a model predictive control algorithm. The model predictive control algorithm calculates the original control instruction by solving an online optimization problem in each control cycle, and the objective function and constraint conditions of the optimization problem are set in real time according to the relevant control parameter configuration set retrieved and loaded by the decision-making module.

[0013] Preferably, the process of the adaptive control module generating the final control signal includes: Receive the original control instruction calculated by the adaptive control module and receive the diagnostic confidence score generated by the intelligent diagnostic module; Based on the received diagnostic confidence score, dynamically determine in real time the key adjustment parameters required for signal smoothing processing in subsequent steps; the determination logic of the key adjustment parameters is configured such that a lower diagnostic confidence score corresponds to a calculated parameter value that can result in a smoother change in the control signal; Perform signal smoothing processing on the original control instruction with the key adjustment parameters determined dynamically in real time; The processed signal is output as the final control signal to the execution interface module.

[0014] Preferably, the multi-sensor adaptive control method for an injection stretch blow molding machine includes: Collect multi-modal process data and perform preprocessing; analyze data from the sensor module, adopt a hierarchical architecture for rapid detection and fine working condition identification, and output diagnostic flags and confidence levels; fuse expert knowledge and online learning results, and determine optimized response strategies according to the confidence levels, and dynamically load a set of control parameters according to the strategies; calculate the original instructions based on the parameter set and data, and generate the final signal by smoothing the instructions according to the confidence levels; apply the final control signal to the actuator of the target device to achieve adaptive control.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The intelligent diagnostic module of the hierarchical architecture, multi-modal fusion and interpretable artificial intelligence adopted by the present invention significantly enhances the ability to extract information from complex sensor data. It can accurately identify specific operating conditions that are difficult to distinguish in the prior art, and provide reliable confidence evaluation and interpretation information, improving the defects of unclear diagnosis and lack of in-depth situation awareness, and providing an accurate and reliable premise for subsequent differential adaptive adjustment.

[0016] 2. The decision-making module that fuses expert knowledge and online learning constructed by the present invention can intelligently convert the accurate information output by the diagnostic module into an optimized response strategy. This mechanism can be dynamically selected according to specific diagnostic results and continuously optimized using historical feedback, overcoming the problems of the adjustment strategy of the existing control system being patternized and lacking differential response capabilities, and ensuring the pertinence and effectiveness of control measures.

[0017] 3. Through dynamic target management, application of the model predictive control algorithm and smooth transition execution of confidence level adjustment, the present invention accurately and stably implements the differential strategies of the upper-level decision-making to the specific control operations of the equipment. The complete, intelligent and adaptive closed-loop from diagnosis, decision-making to execution improves the overall performance of the system in suppressing complex disturbances and ensuring product quality consistency, and generally achieves the goal of intelligent, differential and highly robust adaptive adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic structural diagram of a multi-sensor adaptive control system for an injection stretch blow molding machine according to an embodiment of the present invention; Figure 2 It is a schematic structural diagram of an intelligent diagnostic module according to an embodiment of the present invention; Figure 3 It is a schematic flow diagram of a multi-sensor adaptive control method for an injection stretch blow molding machine according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figures 1 to 3 , the present invention provides the case name, and the technical solution is as follows: Example 1: As an implementation method of the present invention, refer to Figure 1 , the control system of this embodiment includes: The sensor module collects multi-modal process data and performs pre-processing. The sensor module integrates a variety of sensors based on different physical principles, including: The temperature sensor uses an infrared thermal imager and is arranged near the entrance of the heating station and the blow molding station to monitor the temperature distribution on the surface of the bottle blank in real time.

[0021] Pressure sensor, real-time measurement of the axial resistance of the material during the stretching process.

[0022] Laser displacement scanning sensor monitors the geometric profile of preforms and bottles in key processes online.

[0023] Position and speed sensors are used to obtain the real-time kinematic state and synchronization information of the equipment.

[0024] Environmental condition sensors monitor ambient temperature and humidity that affect equipment operation and process stability.

[0025] After collecting the raw multimodal process data from various sensors, the sensor module will perform a series of automatic preprocessing operations to improve data quality, unify data formats and make them suitable for subsequent intelligent diagnosis and control calculations. First, for sensor signals containing noise, a median filter is used to smooth the data, remove random fluctuations and high-frequency interference, and extract clearer signal trends. Secondly, the obvious anomalies and erroneous outliers that may exist in the data stream are detected and processed, and a simple sliding window detection algorithm is used to identify and remove the identified outliers. Next, according to the calibration information of the sensor, the raw electrical signal readings are converted into engineering units with actual physical meaning. In addition, since the sampling frequencies of different sensors may be different, the data streams from different sources need to be timestamped to ensure the synchronization of the data in time. Finally, in order to eliminate the influence of different physical dimensions and numerical ranges on subsequent algorithms, the sensor data is standardized.

[0026] The data management module provides unified data storage and access services for the system. In this embodiment, it is implemented based on the internal storage medium. The data management module is mainly responsible for persistently storing information crucial for the system operation and performing efficient retrieval in response to requests from other modules. The content it stores mainly includes: historical process data and real-time data cache collected and preprocessed by the sensor module; machine learning files and their configurations for the intelligent diagnosis module; expert knowledge base and adaptive response strategy sets for the decision-making module; learning status, intermediate results, and updated strategies and model parameters generated by the online learning mechanism; system operation logs, alarm records, and operator settings.

[0027] Each intelligent function module of the system needs to interact with the data management module to read the required data, models, and knowledge, and write new results and records into the storage.

[0028] The interaction module provides a human-machine interface for the operator to monitor the system status and perform necessary settings. In this embodiment, it is implemented using an industrial touch screen. This module is mainly used to display key operation data processed by the sensor module, system status information, and summary information generated by the intelligent diagnosis module, etc.; at the same time, it is also used to receive basic operation instructions and process parameter setting values input by the operator.

[0029] Intelligent diagnosis module: Analyze the data from the sensor module, and use the machine learning model stored in the data management module to identify the operating conditions, and generate diagnosis identifiers and diagnosis confidence levels. Specifically, the structure diagram of the intelligent diagnosis module is as Figure 2 shown. The intelligent diagnosis module adopts a hierarchical and hierarchical diagnosis architecture, which is divided into a first diagnosis layer and a second diagnosis layer. The first diagnosis layer deploys algorithms with low computational complexity and fast response speed. In this embodiment, a statistical process control method based on key performance indicators is used. The first diagnosis layer judges whether the system deviates from the normal operation mode through the statistical process control method. Specifically, it first calculates the average value and standard deviation of these indicators based on the historical data of key performance indicators collected during the normal operation of the system, and sets the center line and control limits of the statistical control chart accordingly. During the real-time monitoring process, once it is detected that the current value of any KPI exceeds its preset control limit, or the KPI data sequence shows a violation of the preset statistical operation rules, it is determined that the system state has significantly deviated from the normal operation mode, thereby triggering the second diagnosis layer and simultaneously sending a preliminary alarm to the interaction module.

[0030] Further, the second diagnostic layer is activated after receiving the abnormal trigger signal from the first diagnostic layer, and calls the trained machine learning model stored in the data management module. In this embodiment, the system calls a pre-constructed, computationally efficient linearized state-space model based on specific physical knowledge from the data management module. The construction of the model is based on a profound understanding of the heat transfer physical process during the preform heating stage in the injection stretch blow molding machine. Specifically, according to the law of conservation of thermodynamic energy, considering the main physical effects such as the radiative heat transfer of the heating lamps, the absorption and internal heat conduction of the preform material, and the convective heat dissipation between the preform surface and the environment, a nonlinear differential equation describing the temperature change of the key parts of the preform with the heating power and environmental conditions is established. Subsequently, a typical stable operating point is selected, and this nonlinear equation is linearized, and finally a linearized state-space expression form is derived. During operation, this model receives the current power command value of the heating lamps and the readings of the environmental temperature sensors, and real-time predicts the expected temperature dynamic trajectory of several key monitoring points of the preform under ideal heating conditions.

[0031] Further, the actual temperature measurement values of the corresponding key monitoring points of the preform collected by the sensor module are subtracted point by point from the expected temperature dynamic trajectory values predicted by the model in real time, generating a multi-dimensional residual time series vector containing multiple temperature deviation signals. Each element in the vector represents the real-time deviation between the actual temperature of a specific monitoring point and its physical expected temperature, quantifying the degree to which the heating process deviates from the ideal physical model. The system processes the data segment of the multi-dimensional residual time series within a preset time window. For each residual signal in this data segment, the system calculates a set of predefined statistical feature values, including the average value, standard deviation, peak-to-peak value, skewness, kurtosis, zero-crossing rate, and signal energy of the residual signal within the time window. These statistical feature values of all monitoring point residual signals are combined to form a residual feature vector with a fixed length. This feature vector is then input into a gradient boosting decision tree model stored in the data management module and pre-trained offline. The gradient boosting decision tree model has been trained to learn to map specific residual feature vector patterns to specific equipment operating conditions or fault categories. The model efficiently classifies the input residual feature vector by integrating the judgments of multiple decision trees. Table 1 gives examples of some key features and their usually reflected residual patterns.

[0032] Table 1 Predefined statistical feature values and their descriptions Feature Name Calculation Description Main Residual Pattern Reflected Average Value Mean Value of Residual Signal within Time Window Degree of Continuous Bias of Signal Standard Deviation Standard Deviation of Residual Signal within Time Window Fluctuation Amplitude of Signal Peak-to-Peak Value Difference between Maximum and Minimum Values of Signal within Time Window Oscillation or Sudden Jump Amplitude of Signal Skewness Asymmetry of Signal Distribution Imbalance of Signal Deviation Direction Signal Energy Sum of Squares of Residual Signal within Time Window Overall Deviation Intensity of Signal Kurtosis Sharpness of Signal Distribution Frequency Characteristics of Signal Zero-Crossing Rate Number of Times Signal Crosses Zero Line within Time Window Frequency of Extreme Values in Signal Furthermore, the gradient boosting decision tree model outputs a specific and clear operating condition event identifier based on the input residual feature vector. This identifier is one of the predefined category labels, such as problem labels like "decreased aging efficiency of heating lamp tubes", "abnormal preform feeding temperature", "impact of sudden environmental temperature change", or "drift of temperature sensors in a specific area". At the same time, according to the ensemble learning mechanism inside the gradient boosting decision tree model, a diagnostic confidence score is calculated and output based on the voting ratio of all base decision trees for the optimal category, quantifying the credibility of this diagnostic result.

[0033] Furthermore, to enhance the comprehensibility of the diagnostic results and assist in troubleshooting, the intelligent diagnostic module executes an interpretable artificial intelligence analysis process. First, after the gradient boosting decision tree model completes the classification of operating condition events and outputs the diagnostic identifier and confidence level, the system starts the feature impact analysis step. In this step, the system analyzes the trained gradient boosting decision tree model, for example, by calculating the total information gain used for node splitting by each input feature during the construction of the model's decision tree, to quantify the contribution of each input feature to the final classification result. Accordingly, the system identifies and ranks several key residual features that have the most significant impact on the current diagnostic result. Secondly, the feature tracing step is executed. The system associates the key residual features identified in the previous step with their physical sources. Each feature in the residual feature vector is calculated by combining the deviation of specific sensor data from the predicted value of the physical model. The system can trace and determine which specific statistical feature indicators and their corresponding original sensor data sources, such as the standard deviation of the mean residual calculated from the temperature sensor data at a specific location, or the peak-to-peak residual calculated from a certain pressure sensor data, contribute the most to the current diagnostic result. Finally, generating and outputting the explanation information, the intelligent diagnostic module integrates the results of feature impact analysis and feature tracing to form an explanation information report. This report clearly lists the key input features that play a decisive role in this diagnostic conclusion and their corresponding sensor source information. This explanation information is output together with the diagnostic identifier and confidence score, providing intuitive understanding of the fault cause for the operator and providing data support for subsequent equipment maintenance and optimization of control strategies.

[0034] The intelligent diagnosis module effectively addresses the problems in the background technology where the existing system is difficult to deeply understand and distinguish operating conditions, unable to intelligently identify the root causes of faults, and the adaptation adjustment strategy is pattern-based and blind. By precisely quantifying the deviation of key physical quantities from the ideal behavior and performing efficient pattern recognition on these information-rich residual features, this solution can not only achieve fine classification of specific operating conditions and trace the causes, enhancing the interpretability of diagnosis, but also the clear diagnosis identifiers output can drive subsequent decision-making and the decision-making module to execute differentiated and targeted adaptive control, thus significantly improving the optimal performance and long-term operation robustness of the entire system in a complex and changing environment.

[0035] Decision-making module: Receive the diagnosis identifier, combine the expert knowledge and online learning results stored in the data management module, and determine the optimized adaptive response strategy; Dynamically adjust the relevant control parameters for subsequent calculations according to the determined response strategy.

[0036] Specifically, the decision-making module receives in real time the diagnosis identifier and the corresponding diagnosis confidence score output from the intelligent diagnosis module, and uses the received diagnosis identifier to retrieve the predefined reference response strategy that matches this identifier in the expert knowledge base stored in the data management module. This reference strategy represents the standardized processing suggestions for this specific operating condition based on engineering experience and theoretical analysis. The decision-making module first uses the received diagnosis identifier as the search keyword to query and match in the expert knowledge base stored in the data management module. The expert knowledge base pre-stores the standardized processing plans defined by domain experts for various fault diagnosis identifiers in the form of a rule base and a lookup table. After successful matching, the decision-making module extracts the reference response strategy corresponding to the current diagnosis identifier from it. This reference strategy is a standardized processing suggestion set based on engineering experience and theoretical analysis, aiming to quickly stabilize the situation and perform preliminary compensation. For example: If the diagnosis identifier is "a significant decrease in ambient temperature", the corresponding reference strategy is: "Temporarily increase the power setting value of all heating zones by 1% uniformly until the ambient temperature stabilizes".

[0037] Further, the decision-making module takes the received diagnostic identifier as a key representation of the current system state and inputs it into the online learning module also stored in the data management module. In this embodiment, the online learning module is specifically implemented by using a reinforcement learning method based on value iteration. Its operation process is as follows: The online learning module internally maintains a value function Q-value function for evaluating the long-term expected benefits of adopting different coping strategy adjustments in a specific system state. The value function is updated online based on the experience accumulated during the continuous operation of the system. Specifically, after the system executes control according to a certain coping strategy, the online learning module will receive performance feedback information related to the execution effect of this time, such as a reward signal composed of the quantification of product quality, production efficiency, and energy consumption indicators. Using the reward signal and based on the selected reinforcement learning algorithm, the module will update its internal value function, thereby continuously refining the value evaluation of taking different strategy adjustment actions in different states. When the decision-making module requests a recommended strategy from it, the online learning module, based on its currently newly learned value function, determines the strategy adjustment action that can bring the highest long-term expected benefit for the state represented by the input diagnostic identifier. Subsequently, the online learning module converts this internally determined optimal action into a clearly defined output format. For example, it maps it to a set of specific control parameter adjustment suggestions. This result after formatted output is the recommended coping strategy finally provided to the decision-making module for subsequent fusion. This recommended strategy reflects the adjustment plan that the system autonomously learns based on its own operation history and performance feedback and tends to optimize the long-term goal.

[0038] The decision-making module obtains two complementary coping strategies. One is the benchmark solution from the expert knowledge base, representing reliable engineering experience, which ensures that there is always a verified and understandable fallback option available; the other is the recommended solution from the online learning module, reflecting the actual operation effect and potential optimization direction of the equipment, which provides data-driven and continuously adaptable adjustment suggestions. By preparing these two strategy sources simultaneously, the decision-making module has more comprehensive and balanced decision-making information, including both static, rule-based stability considerations and dynamic, performance feedback-based optimization considerations. This provides a necessary and high-quality input premise for subsequent intelligent and risk-controlled fusion judgment, avoiding the one-sidedness or insufficient adaptability that may be brought about by relying solely on a single information source.

[0039] Further, after obtaining the benchmark coping strategy and the recommended coping strategy respectively, the decision-making module executes a strategy fusion step to generate the finally output optimized coping strategy. The fusion process uses a dynamic weighted fusion algorithm based on diagnostic confidence. The core of the algorithm is to dynamically determine the contribution weights of the benchmark coping strategy and the recommended coping strategy in the formation of the final strategy according to the diagnostic confidence score and historical performance feedback.

[0040] Specifically, the decision module first calculates the weight of the recommended strategy and the weight of the benchmark strategy based on the input diagnostic confidence score, ensuring that the sum of these two weight values ​​is always equal to 1, and the weight of the recommended strategy increases monotonically with the diagnostic confidence score through a predefined S-type function. A high confidence score will increase the weight of the recommended strategy, making the decision tend to adopt the optimization result of online learning, while a low confidence score will increase the weight of the benchmark strategy, making the decision tend to be a safe expert plan. The specific formula of the S-type function is as follows:

[0041] in, represents the initial weight of the recommended strategy, Indicates the current diagnostic confidence. is a parameter that controls the steepness of the curve. is the center point of the curve. The value makes the weight Nearby changes are more dramatic, that is, the system is more sensitive to small changes in confidence; smaller A value of 0 makes the transition smoother. It is usually set in the middle area of ​​the confidence score, such as 0.5 or 0.6, to define the turning point of weight distribution. The specific values ​​of these parameters can be adjusted through simulation experiments and combined with expert experience to achieve a reasonable balance between online learning results and expert knowledge at different confidence levels.

[0042] Furthermore, the system continuously tracks and analyzes the sliding window average performance score of the online learning module for a specific diagnostic identifier, for example, calculating performance indicators such as product qualification rate and process efficiency, and comparing this score with the expected benchmark to generate a historical performance adjustment factor. The decision module uses this adjustment factor to adjust the weight, multiplying the initial recommended strategy weight by the historical performance adjustment factor to obtain a temporary adjusted recommended weight, while keeping the initial baseline strategy weight as the temporary baseline weight. To ensure that the final weight sums to 1, the two temporary weights are then normalized, that is, the temporary adjusted recommended weight and the temporary baseline weight are divided by their sum, respectively, to obtain the final recommended strategy weight and the final baseline strategy weight. This final weight reflects both the confidence of the current diagnosis and the consideration of the historical performance of the online learning module. Finally, based on this pair of final weights, the strategy content is integrated, and the adjustment amount of the corresponding parameters in the baseline response strategy and the recommended response strategy is multiplied by their respective final weights, and then the two products are added to obtain the optimized adaptive response strategy.

[0043] The optimized adaptive response strategy calculated through the above dynamic weighted fusion process is an optimized adaptive response strategy that synthesizes expert knowledge, online learning experience and balances risks according to the current diagnostic confidence. The optimized adaptive response strategy is passed to the decision-making module to guide the specific implementation of subsequent control actions.

[0044] The present invention intelligently fuses expert knowledge and online learning results through a decision-making module, and performs dynamic weighting and risk management by using diagnostic confidence and historical performance, overcomes the defects of the traditional adaptive control strategy being patternized and blind, and realizes differential, optimized and robust response strategy decision-making for specific diagnostic working conditions.

[0045] Further, the optimized adaptive response strategy output by the decision-making module will generate a set of relevant control parameters for subsequent adaptive control calculations. Specifically, after the decision-making module outputs a determined response strategy reflected as a specific strategy identifier, based on this identifier, it accesses the information stored in the data management module. The data management module has pre-stored multiple sets of control parameter configuration sets uniquely corresponding to each strategy identifier; the pre-determination of the control parameter configuration sets aims to obtain a set of robust control parameter combinations through systematic offline analysis and optimization for each foreseeable typical working condition, such as working condition problems such as raw material batch change, large environmental temperature fluctuation and initial wear of specific components. These typical working conditions are defined through historical data analysis, expert experience summary and simulation experiments. The decision-making module retrieves and locates the specific parameter configuration set to be applied currently according to the strategy identifier. Each set of configuration sets contains key parameters required for subsequent adaptive control calculations, such as predictions of model predictive control algorithms, target function weights, constraint limits, and underlying controller gains. The parameter configuration sets corresponding to different strategy identifiers are targeted in content. The decision-making module loads the retrieved set of parameter configuration sets as the currently effective configuration, replaces the original settings, and outputs the updated and currently effective set of parameter configuration sets in its entirety to the adaptive control module for it to calculate based on the new parameters in the next control cycle, thus realizing the dynamic, precise adjustment and effective execution of the control strategy.

[0046] The decision-making module retrieves and loads the pre-stored and optimized parameter configuration set from the data management module according to the strategy identifier, realizes the efficient and reliable decoupling between the upper-layer response strategy decision-making and the specific implementation of the underlying control parameters, ensures that different response strategies can be quickly, accurately and consistently converted into the specific parameter configurations required by the downstream calculation module, and thus provides key support and implementation guarantee for the precise execution of differential adaptive control and the effective implementation of high-level decisions.

[0047] The adaptive control module calculates the original control instruction based on the adjusted relevant control parameters and the data from the sensor module; and generates the final control signal that has been smoothed to ensure a stable transition based on the diagnostic confidence and the original control instruction.

[0048] Specifically, the module receives, in each control cycle, the currently effective control parameter configuration set from the decision-making module, the real-time sensor data from the sensor module, and accesses the internal process prediction model stored in the data management module. The model used for internal prediction in MPC selects a linear time-invariant state space model in this embodiment. The state space model is obtained by physically modeling and linearizing the key physical processes of the injection stretch blow molding machine at different operating points. The specific parameters of the model are specified by the currently effective control parameter configuration set. The LTI model is adopted to significantly improve the calculation efficiency to meet the real-time requirement while ensuring sufficient prediction accuracy. To obtain the accurate current state information required by MPC, the module uses a Kalman filter, which fuses the process model defined in step 2 and the real-time sensor measurements to online estimate the key state variables inside the system and provide the optimal state initial value for subsequent prediction and optimization.

[0049] Furthermore, based on the LTI model, the state trajectory of the system within the future prediction time domain is recursively calculated, and the prediction length is set by the parameter set. In each control cycle, an optimization problem is constructed to calculate the optimal control input sequence within the future control time domain. Since the LTI prediction model is adopted and linear constraints are usually set, the optimization problem is transformed into a quadratic programming problem. The optimization objective function (whose weights of each term are set by the parameter set and aims to minimize the tracking error and control energy consumption) and the constraint conditions (whose limits are set by the parameter set, such as actuator saturation and process boundaries) jointly define this QP problem. The system uses an efficient QP solver optimized for real-time embedded applications to online solve this QP problem. From the obtained future optimal control input sequence, the first control action vector is extracted as the original control instruction to be issued in the current control cycle.

[0050] By adopting the model predictive control algorithm and dynamically setting its optimization objective function and constraint conditions according to the parameter configuration set provided by the upstream decision-making module in real time, the adaptive control module can prospectively calculate the optimal original control instruction that can accurately execute the current specific response strategy intention and explicitly meet the system operation limitations, thus ensuring that the high-level adaptive decision can be effectively transformed into high-performance and robust low-level control actions. It is the key computational core to achieve the intelligent and refined control objectives of the entire system.

[0051] Further, the module utilizes a predefined non - linear function to dynamically calculate the core parameter required for the subsequent first - order low - pass filter, the filter time constant, based on the real - time received diagnostic confidence score. The function is designed such that the calculated filter time constant monotonically increases as the diagnostic confidence score decreases. When the diagnostic confidence score is 1, the function outputs a preset minimum time constant; when the diagnostic confidence score is 0, the function outputs a preset maximum time constant; between 0 and 1, the time constant smoothly transitions from the maximum value to the minimum value. The specific form of the function uses an exponential function to achieve the desired transition curve and its maximum / minimum time constant parameters, which are determined through simulation and experiments during the system design and debugging phases, aiming to optimally balance the response speed and stability. The specific calculation formula of the non - linear function is as follows:

[0052] Where represents the filter time constant, represents the preset maximum time constant, represents the preset minimum time constant, represents the diagnostic confidence score.

[0053] Further, after determining the current filter time constant, the module uses a first - order digital low - pass filter to perform real - time filtering on the received original control instruction sequence. The output value of the filter is obtained by weighted averaging the current input and the filter output value at the previous moment through a filtering coefficient. The filtering coefficient is calculated and determined in real - time by the filter time constant and the fixed control sampling period of the system. A larger time constant will calculate a smaller filtering coefficient, making the current filter output more dependent on the output at the previous moment, thus achieving a stronger smoothing effect; conversely, a smaller time constant corresponds to a larger filtering coefficient, allowing the output to follow the input change faster. The signal sequence processed by this first - order digital low - pass filter is the final control signal. The signal ensures the moderation of control actions when the diagnostic uncertainty is high, and the final control signal is then output to the execution interface module.

[0054] Dynamically adjusting the smoothness of the control instruction according to the diagnostic confidence score can automatically enhance the buffer to moderate the control action and ensure system stability when the diagnostic uncertainty is high, and allow a faster control response when the confidence is high. This intelligent transition mechanism significantly improves the robustness and operational safety of the entire adaptive control system under various working conditions.

[0055] The execution interface module receives the final control signal and sends it to the corresponding actuator of the device.

[0056] Specifically, the execution interface module forms a bridge between the entire adaptive control system and the physical actuators of the injection stretch blow molding machine, and is responsible for converting the final control intention generated by the internal calculation of the control system into physical signals that can drive the actual device to act. This module receives the smoothed final control signals output from the adaptive control module. These signals are digital and represent the precise control action states to be executed in the current cycle. The core function of the execution interface module is to perform necessary signal conversion, adaptation, and transmission.

[0057] Table 2 Examples of target actuators and signals processed by the execution interface module Target Actuator Type of Control Signal Required Perform Interface Module Operation Heating Lamp Power Unit Analog Voltage DAC Conversion Output Servo Motor Driver Industrial Fieldbus Message Encode Instruction and Send Bus Data Stepper Motor Driver Pulse / Direction Signal Generate Pulse Sequence with Specified Frequency and Quantity High / Low Pressure Blowing Solenoid Valve Digital Switching Quantity DO (Digital Output) Proportional Flow Valve Analog Current DAC Conversion Output Frequency Converter (Fan / Pump) Modbus Communication Organize and Send Communication Message Table 2 gives specific examples of several common actuators processed by the execution interface module in this embodiment. For the heating lamp power adjustment unit that requires analog control, the digital power command is converted into a standard 0 - 10V analog voltage signal through a digital-to-analog converter (DAC). For the servo motor driver, the digital position, speed, and torque commands are packed into data packets that conform to a specific industrial fieldbus protocol. For switch-type devices such as controlling high-pressure / low-pressure blowing solenoid valves, the logical state is converted into the corresponding digital output (DO) signal.

[0058] Furthermore, while sending the final control signals to the various actuators of the device, the execution interface module also has the ability to monitor the execution status and communication conditions. For example, the execution interface module can capture error codes, execution timeout information, or communication fault signals returned by the actuators (such as servo motor drivers or fieldbus devices). On the one hand, this feedback information from the execution level can be directly alarmed to the operator through the interaction module, indicating abnormalities at the execution end of the device. On the other hand, this information is also transmitted to the data management module for recording and can be used by the intelligent diagnosis module as an auxiliary diagnostic information source to help the system more comprehensively evaluate the device status, distinguish whether it is a process problem or a problem with the actuator itself, thereby indirectly optimizing the subsequent diagnostic accuracy and decision-making rationality. This feedback from the execution interface further enhances the closed-loop characteristics of the system and the perception ability of the actual operating conditions.

[0059] The multi-sensor adaptive control system of the injection stretch blow molding machine proposed by the present invention obtains accurate working condition recognition and confidence through intelligent hierarchical diagnosis combining physical knowledge, effectively improving the problems of unclear diagnosis and difficult identification of the root cause. The decision-making module intelligently integrates expert knowledge and online learning and performs dynamic weighting based on confidence to formulate a differential optimal strategy under controllable risks, overcoming the drawbacks of patternization and blindness in the adaptive adjustment mode. The decision-making module loads a specific parameter set according to the strategy and drives the adaptive control module using model predictive control to accurately execute the optimized and constraint-satisfied control instructions under the intention of the strategy, ensuring the effective implementation of the decision. The adaptive control module smooths the instructions according to the confidence to ensure stable and safe operation. The system completely constructs a closed-loop control process from accurate diagnosis to intelligent decision-making, then to optimized execution and smooth transition, significantly improving the control accuracy, product quality consistency, operation robustness and overall intelligent level of complex industrial processes such as injection stretch blow molding.

[0060] Embodiment 2: As an implementation manner of the present invention, referring to Figure 3 , the control method of this embodiment includes: Collect multi-modal process data and perform preprocessing.

[0061] Continuously collect multi-modal process data during the operation of the injection stretch blow molding machine, covering information such as the temperature distribution in the preform heating zone, the force of the stretching rod, the blowing pressure, the position and speed of key components, and the ambient temperature and humidity. The sensor module performs necessary preprocessing on the collected raw data, such as signal filtering, denoising, calibration, normalization, and timestamp alignment, to form a structured data stream for subsequent analysis.

[0062] Analyze the data from the sensor module, adopt a hierarchical architecture for rapid detection and fine working condition identification, and output diagnostic flags and confidence levels.

[0063] Perform hierarchical diagnosis on the preprocessed data stream. First, the first diagnostic layer inside it adopts the statistical process control (SPC) method based on key performance indicators (KPIs) to continuously monitor whether the main KPIs significantly deviate from the statistical control limits established based on historical normal data. Once the SPC detects an anomaly and determines that the system deviates from the normal operating mode, it immediately generates an anomaly trigger signal and can simultaneously issue a preliminary alarm through the interaction module. After receiving the trigger signal from the first diagnostic layer, the second diagnostic layer of the intelligent diagnostic module is activated. This layer first calls the linearized state-space model stored in the data management module, which is established based on the physical process of preform heating heat transfer, and combines the current device inputs to real-time predict the expected temperature dynamic trajectory of the key monitoring points. Subsequently, subtract the actual temperature measured by the sensor module from this expected trajectory to calculate a multi-dimensional temperature residual time series. Then, the system processes the data of this residual sequence within a preset time window, extracts its statistical features, and forms a residual feature vector. Finally, input this feature vector into the gradient boosting decision tree (GBDT) model that is also stored in the data management module and pre-trained. This GBDT model classifies the residual feature patterns and outputs clear and specific operating condition event identifiers and corresponding diagnostic confidence scores.

[0064] Furthermore, receive the diagnostic identifier and diagnostic confidence score output from the intelligent diagnostic module. The decision-making module executes a policy determination process, retrieves the matching reference response policy in the expert knowledge base stored in the data management module using the diagnostic identifier; at the same time, input the diagnostic identifier into the online learning module in the data management module to obtain the recommended response policy learned based on historical operation data and long-term performance feedback. Then, the decision-making module adopts a dynamic weighted fusion algorithm based on diagnostic confidence and historical performance adjustment factors to integrate the reference policy and the recommended policy. Specifically, calculate the final weight according to the confidence and historical performance, and fuse the adjustment suggestions for the control parameters in the two policies through methods such as weighted average to generate an optimized adaptive response policy.

[0065] Fuse expert knowledge and online learning results, and determine the optimized response policy according to the confidence, and dynamically load the control parameter set according to the policy.

[0066] Receive the optimized adaptive response policy output from the decision-making module. Based on this optimized adaptive response policy, the decision-making module accesses the data management module, retrieves and loads a complete set of pre-stored control parameter configuration sets that are uniquely corresponding to this optimized adaptive response policy. This configuration set contains all the parameters required by the downstream adaptive control module. The decision-making module sets this parameter configuration set as the currently effective configuration.

[0067] Further, receive the currently effective control parameter configuration set from the decision-making module and the real-time data from the sensor module. The module first uses a Kalman filter for state estimation to obtain the accurate current system state. Then, based on the linear time-invariant (LTI) process prediction model stored in the data management module and the current state, predict the future behavior of the system. Next, construct and solve a quadratic programming (QP) optimization problem according to the optimization objective function and constraint conditions set by the current parameter configuration set, and use an efficient real-time QP solver to calculate the optimal control input sequence within the future control time domain. The module extracts the first control action vector of this sequence as the original control instruction for output.

[0068] Calculate the original instruction based on the parameter set and data, and smooth the instruction according to the confidence level to generate the final signal.

[0069] Receive the original control instruction from the adaptive control module and the diagnostic confidence score from the intelligent diagnosis module. The module uses a predefined non-linear function to dynamically calculate the filtering time constant of the first-order digital low-pass filter according to the confidence score. Then, perform real-time low-pass filtering on the original control instruction sequence using the filtering coefficient determined by this dynamic time constant to generate a smoother final control signal that ensures a stable transition.

[0070] Apply the final control signal to the actuator of the target device to achieve adaptive control.

[0071] Receive the final control signal from the adaptive control module. This module is responsible for performing necessary signal conversion and adaptation, and the converted signal is sent to the corresponding physical actuators on the injection stretch blow molding machine to drive them to perform corresponding physical actions, completing the entire adaptive control process.

[0072] Through the cyclic execution of the above steps, this method realizes the in-depth analysis, intelligent diagnosis, adaptive decision-making, and optimal control of multi-sensor data of the injection stretch blow molding machine, can effectively cope with complex working condition changes, and improve production quality and efficiency.

[0073] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Multi-sensor adaptive control system for injection stretch blow molding machine, characterized in that, Including: A sensor module that collects multi-modal process data and performs preprocessing; A data management module that stores and retrieves process data, models, policies, knowledge bases, and learning results required for operation; An interaction module that monitors the system status and receives parameter inputs; An intelligent diagnosis module that analyzes data from the sensor module and uses machine learning models stored in the data management module to identify operating conditions, generating diagnostic identifiers and diagnostic confidence levels; A decision-making module that receives diagnostic identifiers, combines expert knowledge and online learning results stored in the data management module to determine an optimized adaptive response strategy; and dynamically adjusts relevant control parameters for subsequent calculations based on the determined response strategy; An adaptive control module that calculates an original control instruction based on the adjusted relevant control parameters and data from the sensor module; and generates a final control signal that has been smoothed to ensure a stable transition based on the diagnostic confidence level and the original control instruction; An execution interface module that receives the final control signal and sends it to the corresponding actuator of the device.

2. The multi-sensor adaptive control system of the injection stretch blow molding machine according to claim 1, characterized in that The intelligent diagnosis module adopts a hierarchical diagnosis structure including: a first diagnosis layer for rapid anomaly detection and a second diagnosis layer for fine operating condition classification; the intelligent diagnosis module integrates an interpretable artificial intelligence analysis function for providing explanatory information on input features and sensor sources that have a significant impact on the results after diagnosis.

3. The multi-sensor adaptive control system of the injection stretch blow molding machine according to claim 1, characterized in that, The determination of the optimized adaptive response strategy includes: Based on the diagnostic identifier generated by the intelligent diagnosis module, retrieving a benchmark response strategy that matches the current diagnosed operating condition from the expert knowledge base stored in the data management module; Inputting the diagnostic identifier as status information into the online learning module stored in the data management module to obtain a recommended response strategy for the current diagnosed operating condition learned by the online learning module based on historical operation data and long-term performance feedback; Integrating the benchmark response strategy and the recommended response strategy according to a preset fusion algorithm and considering the confidence score of the diagnostic identifier to generate the finally output optimized adaptive response strategy.

4. The multi-sensor adaptive control system of the injection stretch blow molding machine according to claim 3, characterized in that, The preset fusion algorithm adopts a dynamic weighted fusion method, and the determination logic of its weight coefficient includes: initially determining a benchmark weight value according to the current diagnostic confidence score, and generating an adjustment factor using the real-time analysis result of historical performance data, where the adjustment factor is used to adjust the benchmark weight value.

5. The multi-sensor adaptive control system of the injection stretch blow molding machine according to claim 1, characterized in that, The dynamic adjustment implemented by the decision-making module includes: Multiple groups of different control parameter configuration sets corresponding to different response strategy identifiers are pre-stored in the data management module; the decision-making module retrieves and loads the corresponding complete parameter configuration set from the data management module according to the currently received response strategy identifier.

6. The multi-sensor adaptive control system of the injection stretch blow molding machine according to claim 5, characterized in that, The adaptive control module adopts a model predictive control algorithm, and the model predictive control algorithm calculates the original control instruction by solving an online optimization problem in each control cycle, and the objective function and constraint conditions of the optimization problem are set in real time according to the relevant control parameter configuration set retrieved and loaded by the decision-making module.

7. The multi-sensor adaptive control system of the injection stretch blow molding machine according to claim 1, characterized in that The process by which the adaptive control module generates the final control signal includes: Receive the original control instruction calculated by the adaptive control module and receive the diagnostic confidence score generated by the intelligent diagnosis module; Based on the received diagnostic confidence score, dynamically determine in real time the key adjustment parameters required for signal smoothing processing in subsequent steps; the determination logic of the key adjustment parameters is configured such that a lower diagnostic confidence score corresponds to a calculated parameter value that can result in a smoother change in the control signal; Perform signal smoothing processing on the original control instruction using the key adjustment parameters determined dynamically in real time; The processed signal is output as the final control signal to the execution interface module.

8. The multi-sensor adaptive control method for an injection stretch blow molding machine, which adopts the multi-sensor adaptive control system for an injection stretch blow molding machine as described in claim 1, is characterized in that, Comprising: Collect multi-modal process data and perform preprocessing; Analyze the data from the sensor module, perform fast detection and fine working condition identification using a hierarchical architecture, and output a diagnostic flag and confidence; fuse expert knowledge and online learning results, and determine an optimized response strategy based on the confidence, dynamically load a control parameter set according to the strategy; calculate the original instruction based on the parameter set and the data, and generate the final signal by smoothing the instruction according to the confidence; Apply the final control signal to the actuator of the target device to achieve adaptive control.

Images