Toy injection molding compensation control method for thin-walled part warping deformation simulation

Abstract

The present application relates to a toy injection molding compensation control method for thin-walled part warping deformation simulation. By arranging a high time resolution micro pyroelectric sensor array at the node of the main heat disturbance propagation path, real-time collection and preprocessing of multi-point transient temperature change data, extraction of heat propagation fingerprint feature vector, combination of mold material history and service cycle to build a dynamic reference fingerprint template library. Adopting difference detection and adaptive threshold judgment, generating regional activation instructions, calling knowledge graph mapping relationship, formulating regional compensation strategy, loading lightweight compensation algorithm model, driving temperature control unit to realize targeted local heat flow adjustment. The compensation process is monitored and fed back in a closed loop, and the data increment optimization model is optimized, realizing adaptive thermal compensation of the mold throughout its life cycle, improving the local thermal regulation response speed and precision of the mold, effectively prolonging the service life of the mold, and reducing the risk of material thermal degradation.

Description

Technical Field

[0001] This invention relates to the field of mold temperature control coordination and deformation suppression technology in injection molding process, and in particular to a toy injection molding compensation control method for the simulation of warpage deformation of thin-walled parts. Background Technology

[0002] Currently, the injection molding industry for thin-walled toy parts generally faces technical bottlenecks in controlling the geometrical accuracy and stability of products. This is especially true when parts have thin walls and complex shapes, where thermal fluctuations within the mold can easily induce quality defects such as warping and deformation. To address this issue, mainstream industry solutions focus on refined management of mold temperature control systems, multi-point temperature closed-loop control, cooling water path optimization, and global parameter tuning based on classic feedback methods such as PID control. For example, by installing multiple temperature sensors at key locations in the mold, combined with a mold temperature controller to achieve zoned temperature control, and dynamically adjusting actuator power based on the deviation between real-time temperature and setpoint values, efforts are made to minimize thermal distribution fluctuations on the cavity surface. With the development of intelligent manufacturing and the Industrial Internet of Things, some recent research trends are attempting to introduce digital twins, physical field simulation, and machine learning-assisted hotspot identification and compensation optimization algorithms to achieve predictive control of the mold temperature field in the spatiotemporal dimensions.

[0003] However, the existing technical solutions still have several prominent limitations in practical applications. First, the disturbance identification and compensation strategies of most systems rely heavily on static temperature threshold criteria: the relevant control actions are only triggered once the actual detected temperature value exceeds the preset upper or lower limit. This reactive response mode means that when rapid disturbances occur in the local thermal state of the mold, the controller often lags behind the physical process, and by the time the compensation action is initiated, the product cavity has already formed irreversible initial deformation, making it difficult to suppress warping in time. Second, common compensation control mainly relies on global parameter tuning, which increases signal processing and calculation delays and may also cause "crosstalk" effects with unaffected areas, making it difficult to achieve accurate management of local short-cycle fluctuations. Furthermore, traditional water circuit modeling and temperature control logic often require complex modeling and pre-tuning of the mold structure and cooling pipes, making it difficult to dynamically adapt to irreversible changes such as mold material aging and thermal property deterioration, resulting in a significant decrease in compensation effect after long-term operation. In addition, most existing research focuses on large or thick-walled products and lacks the ability to detect and respond to asynchronous local deformation of thin-walled molds under high-frequency, micro-amplitude thermal fluctuation environments.

[0004] As the high-end toy industry increasingly demands consistency and automated production cycle time for injection-molded thin-walled parts, overcoming the technical bottlenecks of traditional temperature threshold triggering and delayed judgment when local thermal disturbances occur in the mold, and rapidly and synchronously sensing the initiation of disturbances and proactively implementing effective regional compensation during their propagation, has become an urgent industry challenge. In particular, addressing the issues of high warpage rates, difficulty in ensuring batch consistency, and production cycle time delays caused by the response delays of traditional compensation strategies, the technology urgently needs innovative disturbance-driven, signal-pre-positioned, and regional compensation control methods to achieve real-time sensing and synchronous decision-making of the mold's thermal dynamic evolution process, fundamentally improving molding quality and automation levels.

[0005] This invention proposes a new paradigm for dynamic compensation based on thermal disturbance propagation fingerprints. It utilizes the three-dimensional dynamic response characteristics of mold materials and structures to disturbance signals to achieve highly sensitive and low-latency identification of disturbance initiation. This is then used as a trigger to quickly drive the regional compensation path, thereby eliminating the bottleneck in warping deformation control caused by the lag in disturbance identification. This provides a precise, efficient, and adaptive new paradigm for the mass production of thin-walled parts. Summary of the Invention

[0006] This application provides a toy injection molding compensation control method for the simulation of warpage deformation of thin-walled parts, aiming to solve one of the problems or issues of the prior art mentioned in the background art above.

[0007] The toy injection molding compensation and control method for simulated warpage deformation of thin-walled parts provided in this application specifically includes: S1: Deploy a high time-resolution micro pyroelectric sensor array in the key thermally sensitive area of ​​the mold to acquire the original thermal response signal sequence containing multiple tags at different locations; S2: Analyze the original thermal response signal sequence to obtain the thermal propagation fingerprint feature vector; S3: Construct a benchmark fingerprint template library based on thermal propagation fingerprint feature vectors; S4: Calculate the difference between the current thermal propagation fingerprint feature vector and the corresponding template in the benchmark fingerprint template library in real time, and make a logical judgment between the difference and the preset dynamic threshold that is adaptively adjusted according to the thermal history of the mold to generate a region activation command that identifies the local thermal disturbance start state. S5: Based on the region activation instruction, invoke the perturbation propagation mapping relationship pre-stored in the local knowledge graph to generate a regionalized compensation strategy parameter set for a specific perturbation mode; S6: Based on the regional compensation strategy parameter set, load the compensation algorithm sub-model containing only adjustable parameters, and perform short-term response calculation to generate the region-specific compensation control quantity after regional mapping compression; S7: Drive the temperature control unit that is strongly coupled with the thermal disturbance influence domain according to the region-specific compensation control quantity, while keeping the other unrelated actuators in a silent state, to complete the coordinated adjustment of the local thermal distribution of the mold.

[0008] The toy injection molding compensation and control method for warpage deformation simulation of thin-walled parts provided in this application has the following beneficial effects: (1) To address the problem that traditional injection mold temperature control systems rely on static temperature thresholds for compensation and suffer from response lag leading to warpage control failure, this solution proposes a novel compensation triggering mechanism based on the propagation dynamics of thermal disturbances in the mold steel body for early identification and regionalized response. Existing technologies typically initiate regulation only after the temperature exceeds the limit or deformation has occurred, making it difficult to cope with transient instability caused by rapid thermal fluctuations, especially in high-precision molding scenarios such as thin-walled toy parts, which can easily cause irreversible defects. This invention, by deploying a high-temporal-resolution micro pyroelectric sensor array in key thermally sensitive areas and scientifically arranging the nodes of the main propagation path of thermal disturbances determined by three-dimensional thermal conduction simulation, breaks through the traditional approach of only covering high-temperature areas or surfaces, and achieves accurate capture of the initial excitation and propagation process of thermal disturbances. By combining time-frequency joint analysis to extract the rise time, peak attenuation rate, and phase shift to form a "thermal propagation fingerprint," the system can identify the initiation trend of local thermal anomalies before the temperature deviates significantly from the set value, thereby moving the compensation decision forward to the disturbance initiation stage, significantly improving the predictability and timeliness of the control system.

[0009] (2) To address the thermal response drift caused by material aging, differences in heat treatment status, and extended service life during mold service, this solution constructs a three-dimensional calibration library of mold material batches, heat treatment status, and service life. This library enables normalized comparison of thermal propagation fingerprints at the same physical location under different lifecycles and dynamically updates the benchmark template to ensure the long-term effectiveness of the criteria. During online operation, the system calculates the difference between the current fingerprint and the benchmark template in real time. When the difference exceeds the dynamic threshold adaptively adjusted by thermal history, it is determined to be the start of a predictable disturbance event, and an "area activation command" is immediately issued. This mechanism eliminates the reliance on steady-state deviation or deformation inference in traditional methods and avoids response delays caused by waiting for the temperature to exceed the limit. Furthermore, the system introduces historical disturbance mapping relationships stored in a local knowledge graph to accurately associate specific fingerprint anomalies with their corresponding spatial influence domain, expected deformation direction, and optimal compensation action point, guiding the rapid invocation of the lightweight compensation sub-model. This sub-model has simplified parameters (only 3–5 adjustable parameters) and is dedicated to short-term responses, avoiding the computational burden caused by full model retraining and ensuring the real-time performance and stability of control command generation.

[0010] (3) At the execution level, this scheme abandons the global PID tuning mode and instead implements a local drive strategy based on spatial coupling strength screening: only 2-3 temperature control units (such as microchannel solenoid valves and pulsed Peltier modules) that are strongly correlated with the disturbance influence domain are activated, while the remaining units remain silent. This reduces the inertial interference caused by multi-actuator collaboration and makes full use of existing hardware resources to achieve fine-grained regional control. This "perception-judgment-mapping-response" closed-loop system reconstructs the mold itself into an active signal generator with self-sensing capabilities. It utilizes its inherent heat propagation characteristics as a natural sensing medium to achieve decoupling and pre-positioning of disturbance identification and compensation actions in the time dimension. Compared with traditional methods based on complex waterway modeling or offline simulation pre-compensation, this approach does not require precise flow channel parameter modeling, nor does it rely on the accumulation of a large amount of trial and error data, thus possessing stronger engineering applicability and scalability. Overall, without increasing hardware investment, the system significantly improves the response sensitivity, control accuracy and energy efficiency to local thermal disturbances, effectively suppresses warping deformation in thin-walled part forming, and is especially suitable for intelligent manufacturing scenarios of high-cycle, multi-variable toy products.

[0011] The aforementioned technical approaches collectively construct a closed-loop adaptive control system for early identification and low-latency area compensation of mold thermal disturbances. This system achieves a paradigm shift from "passive temperature control" to "active sensing + intelligent mapping + precise intervention," fundamentally overcoming the bottlenecks of traditional solutions such as lag response, coarse control, and poor adaptability. It provides a reliable guarantee for the quality stability and production efficiency of the precision injection molding process. Attached Figure Description

[0012] Figure 1 This is the main flowchart of a toy injection molding compensation control method designed for warpage deformation of thin-walled parts.

[0013] Figure 2 This is a sub-flowchart of a toy injection molding compensation control method designed for warpage deformation of thin-walled parts.

[0014] Figure 3 This is another sub-flowchart of a toy injection molding compensation control method designed for warpage deformation of thin-walled parts. Detailed Implementation

[0015] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0016] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0017] like Figure 1 As shown, this application provides a toy injection molding compensation and control method for detecting warpage deformation of thin-walled parts, specifically including: S1: Deploy a high time-resolution micro pyroelectric sensor array in the key thermally sensitive area of ​​the mold to acquire the original thermal response signal sequence containing multiple tags at different locations; S2: Analyze the original thermal response signal sequence to obtain the thermal propagation fingerprint feature vector; S3: Construct a benchmark fingerprint template library based on thermal propagation fingerprint feature vectors; S4: Calculate the difference between the current thermal propagation fingerprint feature vector and the corresponding template in the benchmark fingerprint template library in real time, and make a logical judgment between the difference and the preset dynamic threshold that is adaptively adjusted according to the thermal history of the mold to generate a region activation command that identifies the local thermal disturbance start state. S5: Based on the region activation instruction, invoke the perturbation propagation mapping relationship pre-stored in the local knowledge graph to generate a regionalized compensation strategy parameter set for a specific perturbation mode; S6: Based on the regional compensation strategy parameter set, load the compensation algorithm sub-model containing only adjustable parameters, and perform short-term response calculation to generate the region-specific compensation control quantity after regional mapping compression; S7: Drive the temperature control unit that is strongly coupled with the thermal disturbance influence domain according to the region-specific compensation control quantity, while keeping the other unrelated actuators in a silent state, to complete the coordinated adjustment of the local thermal distribution of the mold.

[0018] Step S1: Deploy a high-time-resolution micro pyroelectric sensor array in the critical heat-sensitive area of ​​the mold to acquire the original thermal response signal sequence containing tags at multiple different locations. Specifically, this includes: S1.1: Based on the distribution information of the main path nodes of thermal disturbance propagation identified by the three-dimensional thermal conduction simulation model of the mold, the thermally sensitive area of ​​the mold steel structure is divided to determine the optimal spatial layout coordinate set of the micro pyroelectric sensor array on the mold surface, and generate a mapping table of thermally sensitive area location labels with physical meaning.

[0019] Load the three-dimensional heat conduction simulation model of the mold, extract the transient temperature field gradient distribution data during the injection cycle, and identify areas where the rate of change of heat flux density exceeds a preset threshold as potential heat-sensitive areas.

[0020] The mesh is refined for potential heat-sensitive areas. The propagation time delay matrix of thermal disturbance from the gate to the cavity edge is calculated based on finite element analysis, and the coordinates of key nodes on the propagation path are selected.

[0021] Based on the spatial topology of the key node coordinates, the K-means clustering algorithm is used to discretize the continuous space into several independent heat-sensitive sub-regions, ensuring that the heat conduction characteristics of each sub-region are homogeneous.

[0022] Calculate the geometric center and boundary extreme points of each heat-sensitive sub-region, and combine the mold cooling water channel layout constraints to eliminate interference areas where sensors cannot be installed, generating a set of candidate placement points.

[0023] Sensitivity analysis was performed on the candidate deployment point set to evaluate the response gain of each point to thermal changes in the adjacent area. The point set with the largest response gain and the best spatial coverage was selected as the final deployment coordinates.

[0024] Assign a unique location identifier to each final deployment coordinate and record its three-dimensional spatial coordinate value relative to the mold reference angle, and construct a thermal sensitive area location label mapping table containing physical location information and logical labels.

[0025] Through the above processing method, the simulation data from the previous step is transformed into an optimal set of sensor deployment coordinates and a location label mapping table with clear physical meaning, thereby achieving accurate coverage of the main path nodes of thermal disturbance propagation and establishing a sensing foundation.

[0026] S1.2: Based on the location tag mapping table of the heat-sensitive area, install a high time resolution micro pyroelectric sensor array at the determined optimal spatial layout coordinates, and configure the sampling clock synchronization mechanism of each sensor to set a sampling interval parameter of less than or equal to ten milliseconds, thereby constructing a multi-point distributed thermal sensing hardware network with a unified time base.

[0027] S1.3: The multi-point distributed thermal sensing hardware network is used to collect transient temperature change data in real time during the injection molding process. The analog voltage signal output by each sensor is converted from analog to digital and timestamped to generate a discrete original thermal response signal sequence with a unique location tag and strict time synchronization.

[0028] It receives analog voltage signals from each miniature pyroelectric sensor in a multi-point distributed thermal sensing hardware network. These signals represent a continuous physical quantity characterizing the local transient temperature change of the mold.

[0029] A high-precision analog-to-digital converter is used to discretize the analog voltage signal. Based on the preset ten-millisecond sampling interval parameter, the temperature fluctuation in the continuous time domain is converted into a digital sequence to ensure that the time resolution of the data meets the sub-period level response requirements.

[0030] Each sensor channel is assigned a unique physical location identifier, which corresponds strictly to the thermal sensitive area location label mapping table generated in S1.1, thus realizing the logical binding between the data source and spatial coordinates.

[0031] A global synchronization clock signal is introduced as a time reference to stamp the data frames acquired by each channel with a uniform precision, eliminating the phase deviation caused by multi-channel parallel acquisition and ensuring strict alignment of all node data on the time axis.

[0032] Digital signals with location tags and timestamps are reassembled into a structured data stream, forming a discrete original thermal response signal sequence containing spatial dimension indexes and time dimension sequences.

[0033] Through the analog-to-digital conversion and spatiotemporal alignment processing described above, the output of the hardware network established in the previous step is transformed into digital basic data with spatiotemporal consistency, realizing the accurate conversion from analog physical signals to computable discrete data sequences, and providing high-fidelity input for subsequent fingerprint feature extraction.

[0034] For example, for eight miniature pyroelectric sensors deployed near the mold gate and in the distant cooling channels, a 16-bit resolution ADC is used to synchronously acquire voltage signals at a frequency of 100Hz (i.e., 10ms interval). Channels 1 to 8 are respectively assigned IDs. 01 To ID 08 Location tag. ID was collected during the injection molding holding pressure stage. 03 The sensor voltage at t=1.2s is 0.45V, which is quantized and converted into a digital value of 29491, and then marked with a timestamp T. global =1200ms. Data from all 8 channels is locked under the same global clock pulse, generating a matrix structure containing 8 columns of data, each column's length increasing over time. This process ensures that minute differences (e.g., on the order of microseconds) in the arrival time of thermal disturbances at different locations are accurately recorded, avoiding phase offset calculation errors caused by clock asynchrony, and significantly improving the accuracy of phase feature calculation in subsequent thermal propagation fingerprints.

[0035] S1.4: Perform sliding window denoising and baseline drift correction on the discretized original thermal response signal sequence to remove noise components introduced by environmental electromagnetic interference and sensor zero-point drift, and output a pre-processed high signal-to-noise ratio standardized original thermal response signal sequence as the direct data source for subsequent time-frequency joint analysis.

[0036] Receive a discrete raw thermal response signal sequence with a unique location tag and strict time synchronization. This sequence includes environmental electromagnetic interference and sensor zero-point drift noise.

[0037] The discretized original thermal response signal sequence is subjected to sliding window mid-range filtering. The window width is set to an odd multiple of the number of sampling points to cover the high-frequency pulse interference period, and transient spike noise caused by the start and stop of the injection molding machine servo motor is eliminated, resulting in a preliminarily smooth thermal response data stream.

[0038] Based on the preliminary smoothed thermal response data stream, a baseline drift model is constructed using a polynomial fitting method. A low-order polynomial function is selected to approximate the slow zero-point offset trend of the sensor during long-term operation, and the fitting curve characterizing the amount of baseline drift is calculated.

[0039] By subtracting the fitted curve representing the baseline drift from the initially smoothed thermal response data stream point by point, the DC component offset caused by the overall temperature rise of the mold is eliminated, and a zero-mean thermal response signal with baseline drift removed is obtained.

[0040] Standard deviation normalization is performed on the zero-mean thermal response signal after removing baseline drift. The local standard deviation of the signal within the sliding window is calculated, and the amplitude of each sampling point is divided by this local standard value to eliminate the inconsistency of dimensions caused by the sensitivity differences of different sensors.

[0041] By using sliding window denoising and baseline drift correction methods, the discretized original thermal response signal sequence from the previous step is transformed into a preprocessed, high signal-to-noise ratio standardized original thermal response signal sequence, achieving the technical effects of noise suppression and benchmark unification, and serving as a direct data source for subsequent time-frequency joint analysis.

[0042] For example, for a thermal response signal sequence with a sampling frequency of 100Hz, a sliding window width of 21 sampling points is set, and median filtering is performed to remove spike noise with an amplitude exceeding 3 times the standard deviation. A third-order polynomial is used to fit the baseline drift, the fitted curve is calculated and subtracted from the signal, eliminating a linear temperature rise drift of approximately 0.5℃ / min. The local standard deviation of the remaining signal is calculated and Z-score normalized to unify the variance of each channel signal to 1. After processing, the signal-to-noise ratio of the signal is improved from 15dB to 35dB, and the baseline drift residual is less than 0.01℃, meeting the requirement of input signal stationarity for subsequent wavelet packet decomposition.

[0043] S1.5: The high signal-to-noise ratio normalized original thermal response signal sequence is reorganized into a multidimensional spatiotemporal data matrix according to the spatial topological relationship, and a complete multidimensional sensing data foundation covering the main path nodes of thermal disturbance propagation is established to ensure that subsequent steps can accurately extract the thermal propagation fingerprint feature vector characterizing the local thermal dynamics.

[0044] Step S2: Analyze the original thermal response signal sequence to obtain the thermal propagation fingerprint feature vector. Specifically, this includes: S2.1: Wavelet packet decomposition is performed on the original thermal response signal sequence containing multiple different location labels. Orthogonal wavelet basis functions are used to decompose the non-stationary heat flow signal into multi-scale frequency band sub-signals to obtain a time-frequency domain decomposition coefficient matrix that covers high-frequency transient change information and low-frequency steady-state trend information, thereby completing the first-level data conversion from the original thermal response signal sequence to the time-frequency domain decomposition coefficient matrix.

[0045] Receive a preprocessed, high signal-to-noise ratio normalized raw thermal response signal sequence containing multiple spatial location labels and tightly aligned timestamp data.

[0046] Discrete wavelet packet transform is performed on the signal at each independent location. The db4 orthogonal basis function in the Daubechies wavelet family is selected as the decomposition kernel to match the transient change characteristics of the heat conduction signal of the injection mold steel body.

[0047] The decomposition layer is set to 3 layers, and the original time-domain signal is recursively decomposed into 8 frequency band sub-nodes to ensure that the frequency resolution covers the entire spectrum from low-frequency steady-state trend to high-frequency noise interference.

[0048] Calculate the wavelet packet coefficients of each decomposition node and construct a three-dimensional time-frequency domain decomposition coefficient matrix, where the first dimension is the time index, the second dimension is the sensor spatial location ID, and the third dimension is the frequency band node index.

[0049] By utilizing the orthogonal transformation properties, the non-stationary heat flow fluctuation energy is dispersed into different time-frequency atoms, thereby achieving a joint sparse representation of the signal in both the time and frequency domains.

[0050] By utilizing the localization properties of wavelet coefficients, high-frequency detail coefficients reflecting the arrival of the thermal perturbation front are preserved, while low-frequency approximation coefficients characterizing the background temperature field are separated.

[0051] By using wavelet packet decomposition, the time-domain signal from the previous step is transformed into a time-frequency domain decomposition coefficient matrix with multi-scale resolution, thus achieving the desired technical effect of refined decomposition and structured expression of the characteristics of non-stationary thermal signals.

[0052] For example, for a cavity center sensor signal with a sampling frequency of 100Hz, a 3-level wavelet packet decomposition was performed using the db4 wavelet basis. The original signal length was 1024 points, which were decomposed into 8 sub-band coefficient sequences, each with a length of 128 points. The calculation showed that the 5th sub-band (corresponding to the 12.5-25Hz frequency band) had the highest energy proportion, reaching 65% of the total energy. This frequency band mainly carries the information of rapid thermal shock caused by the injection of the molten metal. The time-frequency domain decomposition coefficient matrix has a dimension of 128×1×8, completely recording the transient energy distribution at this location in a specific frequency band, providing an accurate data foundation for subsequent selection of characteristic frequency bands.

[0053] S2.2: Based on the time-frequency domain decomposition coefficient matrix, perform energy entropy calculation and reconstruction method, screen out the characteristic frequency bands carrying the main energy distribution of thermal disturbance and reconstruct to obtain the denoised pure thermal transient waveform curve, so as to eliminate sensor noise and environmental clutter interference, obtain a pure thermal transient waveform curve with significantly improved signal-to-noise ratio, and realize the second-level data conversion from the time-frequency domain decomposition coefficient matrix to the pure thermal transient waveform curve.

[0054] Based on the time-frequency domain decomposition coefficient matrix generated by S2.1, the energy spectral density of each frequency band sub-signal is calculated to quantify the energy distribution ratio of different frequency components during the thermal disturbance transient process. A frequency band energy entropy evaluation function is constructed based on Shannon's information entropy theory, and the normalized energy entropy value of the j-th frequency band is calculated using the following formula: in, Let the energy entropy of the j-th frequency band be... Let N represent the energy probability distribution at time point i in frequency band j, and N be the number of sampling points. This index is used to characterize the orderliness and complexity of a signal within a specific frequency band.

[0055] Feature frequency bands with energy entropy values ​​below a preset sparsity threshold were selected. These bands primarily carry the transient energy of thermal conduction in the mold steel, while high-entropy bands mainly correspond to random noise and environmental clutter. Wavelet packet reconstruction was performed on the coefficients of the selected feature frequency bands, using inverse wavelet transform to map the frequency domain coefficients back to the time domain. A soft-threshold denoising mechanism was introduced during the reconstruction process, setting coefficients below the noise estimation level to zero to further suppress residual high-frequency interference. The reconstructed time-domain waveforms of each feature frequency band were linearly superimposed to synthesize a denoised, clean thermal transient waveform curve. This curve retains the key morphological characteristics of thermal disturbance propagation while significantly improving the signal-to-noise ratio.

[0056] Through the above-mentioned energy entropy screening and reconstruction process, the time-frequency domain decomposition coefficient matrix of the previous step is transformed into high-fidelity pure thermal transient waveform curve data, realizing the second-level data conversion from noisy mixed signal to key dynamic feature carrier, laying a high-quality data foundation for subsequent accurate extraction of rise time and decay rate parameters.

[0057] For example, the energy entropy of each node in the 8-layer wavelet packet coefficient matrix obtained from the decomposition is calculated. An energy entropy threshold of 0.65 is set, and three characteristic frequency bands with lower entropy values ​​(corresponding to a frequency range of 5Hz-50Hz) are selected. Daubechies-4 inverse wavelet transform is performed on the coefficients of these three frequency bands for reconstruction, and a general thresholding rule is applied for soft thresholding, with the threshold set to three times the noise standard deviation. The superimposed reconstructed waveform shows that the signal-to-noise ratio is increased from 12dB to 28dB, effectively eliminating high-frequency glitches introduced by the vibration of the injection molding machine's hydraulic system. The output clean thermal transient waveform curve is smooth with clear inflection points, meeting the microsecond-level edge detection accuracy requirements.

[0058] S2.3: Perform step response edge detection processing on the pure thermal transient waveform curve, use the adaptive slope threshold method to lock the waveform inflection point at the arrival time of thermal disturbance, accurately measure the time span required for the temperature to rise from the reference temperature level to 90% of the peak temperature, and extract the rise time parameter that characterizes the speed of heat conduction to complete the third-level data conversion from the pure thermal transient waveform curve to the rise time parameter.

[0059] The pure thermal transient waveform curve obtained by the reconstruction in step S2.2 is received. This curve has eliminated high-frequency noise and retained the main energy characteristics of thermal disturbance propagation, and is used as the direct input object for the extraction of rise time parameters.

[0060] The baseline temperature level is locked on the pure thermal transient waveform curve. The average temperature sampling point within the stabilization time window before the thermal disturbance arrives is selected and established as the reference temperature zero point to eliminate the interference of ambient temperature fluctuations on the initial judgment.

[0061] An adaptive slope threshold method is used to perform edge detection on the rising phase of the waveform. The differential slope sequence of adjacent sampling points within the sliding window is calculated. When the slope value exceeds the preset dynamic sensitivity threshold, it is determined to be the arrival time of the thermal disturbance front, and the timestamp of that time is recorded.

[0062] The waveform is tracked from the reference temperature to the peak temperature. The ratio of the current temperature value to the steady-state peak temperature is monitored in real time. When the temperature rises to 90% of the peak temperature, the timestamp corresponding to the critical point is accurately captured.

[0063] A mathematical model for the rise time parameter is constructed, and the key indicators characterizing the rate of heat conduction are calculated using the following formula: in, The rising edge time parameter, To reach 90% of the peak temperature, This represents the arrival time of the thermal disturbance front.

[0064] The calculated rise time parameter is bound to the sensor location label to form a temporal feature component with spatial attributes, which is used to construct a multidimensional thermal propagation fingerprint vector.

[0065] By using the above-mentioned step response edge detection and precise time span measurement processing method, the pure thermal transient waveform curve of the previous step is transformed into the rising edge time parameter of the quantified thermal conduction dynamic characteristics, realizing a high-precision characterization of the propagation speed of local thermal disturbances, and providing a sub-millisecond response triggering basis for compensation control strategies.

[0066] S2.4: Based on the pure thermal transient waveform curve, perform exponential fitting decay analysis, construct a nonlinear regression model of the heat release process and solve its time constant, calculate the temperature drop rate per unit time after the peak temperature, so as to extract the peak decay rate parameter characterizing the heat dissipation efficiency and heat capacity characteristics of the mold material, and realize the fourth-level data conversion from the pure thermal transient waveform curve to the peak decay rate parameter.

[0067] The system receives the clean thermal transient waveform curve reconstructed in step S2.2. This curve has eliminated high-frequency noise and retained the main energy distribution characteristics of thermal disturbance. The data segment after the peak temperature point in the clean thermal transient waveform curve is truncated, and the zero time point is set as the peak time to construct a heat release dataset describing the natural cooling process of the mold steel.

[0068] A first-order thermal relaxation model characterizing the temperature change over time was established by performing exponential decay fitting on the heat release dataset using the nonlinear least squares method. This model is based on a discretized form of Newton's law of cooling and assumes that the local heat capacity and heat dissipation coefficient remain constant over short time periods.

[0069] Construct the following exponential decay function model: in, The temperature value at time t. Based on ambient temperature, Peak temperature The thermal time constant to be solved represents the magnitude of the thermal inertia of the mold material.

[0070] The Levenberg-Marquardt optimization algorithm was used to iteratively solve the time constant in the above equations. By minimizing the sum of squared residuals between the measured temperature series and the model predictions, the optimal fitting parameters were obtained. The time constant directly reflects the combined rate of heat diffusion from the sensor contact point to the mold substrate and transfer to the cooling medium.

[0071] Based on the time constant obtained from the solution, the rate of temperature decrease per unit time is calculated, i.e., the peak decay rate parameter. This parameter is defined as the reciprocal of the time required for the temperature to decay to 36.8% (i.e., 1 / e) of the difference between the peak and the reference temperature.

[0072] Peak decay rate quantifies the heat dissipation efficiency of a local area after thermal disturbance and is a key fingerprint feature for identifying changes in heat capacity caused by microcracks, air gaps, or blockages in cooling water channels inside the mold.

[0073] The calculated peak decay rate parameter is concatenated with the rise time parameter extracted in step S2.3 to form a two-dimensional dynamic feature subset that includes heat conduction rate and heat dissipation efficiency.

[0074] By using exponential fitting decay analysis and time constant solution, the pure thermal transient waveform curve from the previous step is transformed into a peak decay rate parameter that characterizes the heat dissipation efficiency and thermal capacity of the mold material. This achieves the fourth-level data transformation from time-domain waveform to thermal property fingerprint characteristics, providing core dimensional support for the subsequent construction of a highly discriminative thermal propagation fingerprint vector.

[0075] S2.5: Perform cross-correlation calculation on the rising edge time parameter and peak attenuation rate parameter output from adjacent spatial nodes. Determine the time delay of signal propagation in the mold steel medium by calculating the extreme point position of the cross-correlation function. Combine the sampling frequency to convert it into an angle difference value to extract the phase offset relative to the neighboring sensor, which characterizes the spatial propagation directionality and consistency of thermal disturbance. Finally, integrate the rising edge time parameter, peak attenuation rate parameter and phase offset relative to the neighboring sensor to generate a thermal propagation fingerprint feature vector, completing the fifth level of data conversion from multi-dimensional time series parameters to thermal propagation fingerprint feature vector.

[0076] The rise time parameter and peak decay rate parameter output by adjacent spatial nodes are received to construct a cross-correlation calculation model based on the topology of the mold steel medium.

[0077] Discrete cross-correlation function calculations are performed on the pure thermal transient waveform curve to quantify the relationship between the similarity of the two sensor signals in the time domain and the time delay.

[0078] Traverse the preset time delay search interval to locate the global maximum point of the cross-correlation function. The time lag corresponding to this extreme point is the physical time difference of thermal disturbance propagation between the two points.

[0079] By combining the high sampling frequency parameters set by the system, the time delay is converted into a phase angle difference, which characterizes the direction vector characteristics of the thermal wave front propagating inside the mold.

[0080] By integrating the extracted rise time, peak attenuation rate, and phase offset relative to neighboring sensors, a three-dimensional feature vector space is constructed.

[0081] By using a multi-dimensional feature fusion processing method, the time-series parameters of the previous step are transformed into thermal propagation fingerprint feature vectors that characterize local thermal dynamics. This achieves a technological leap from single temperature monitoring to the perception of thermal disturbance propagation mechanisms, providing high-precision data support for subsequent disturbance source localization.

[0082] like Figure 2 As shown, step S3 involves constructing a baseline fingerprint template library based on the thermal propagation fingerprint feature vector. Specifically, this includes: S3.1: Obtain a three-dimensional calibration information dataset containing mold material batch number, heat treatment process parameter curve, and cumulative service cycle duration. Use a multi-dimensional index mapping method to perform spatiotemporal association binding processing on the three-dimensional calibration information dataset and the heat propagation fingerprint feature vector generated in the previous main step to generate a labeled heat propagation fingerprint feature vector sequence with complete history tags, thereby completing the first-level data transformation from independent physical parameters to a labeled heat propagation fingerprint feature vector sequence with integrated history information.

[0083] The system receives the thermal propagation fingerprint feature vector generated in step S2, which includes rise time, peak decay rate, and phase offset. Simultaneously, it retrieves the current mold's material batch number, heat treatment process parameter curve, and cumulative service life from the mold lifecycle management database. A three-dimensional calibration information dataset is constructed, using the material batch as a static attribute index, the heat treatment curve as an initial state constraint, and the service life as a dynamic evolution variable, forming a complete history label describing the mold's physical state.

[0084] A multidimensional index mapping method is employed, using the sensor's spatial location coordinates as the key, to establish a spatiotemporal association between the thermal propagation fingerprint feature vector and the three-dimensional calibration information dataset. A hash mapping mechanism ensures that the thermal response characteristics of each physical location uniquely correspond to a specific set of material and service state parameters, eliminating matching ambiguities caused by data heterogeneity.

[0085] Data fusion processing is performed to embed static material thermophysical parameters and dynamic service aging factors into the metadata field of the thermal propagation fingerprint feature vector. Linear interpolation is used to smooth and complete the discontinuously acquired heat treatment process parameters, ensuring the continuity of calibration information over time, and generating a tagged thermal propagation fingerprint feature vector sequence with complete historical labels.

[0086] The labeled sequences undergo consistency verification to remove missing and outlier values ​​caused by sensor malfunctions or communication interruptions, ensuring the integrity and reliability of the input data. Through this processing, the results of the previous step are transformed into a labeled thermal propagation fingerprint feature vector sequence incorporating historical information. This achieves the first-level data transformation from independent physical parameters to a multi-dimensional coupled state representation, providing standardized input for subsequent staged clustering analysis.

[0087] For example, for a toy thin-walled injection mold made of P20 mold steel, with material batch number P20-2023-A05, the heat treatment process shows a quenching temperature of 840℃ and a tempering temperature of 520℃, and a current cumulative service life of 15,000 cycles. The system extracts sensor data located near the gate at coordinates (120, 80, 0), whose thermal propagation fingerprint feature vector is [rise time: 12.5ms, peak decay rate: 0.85 / s, phase offset: 15°]. The multidimensional index mapping method binds this vector to the material's thermal conductivity of 42 W / (m·K) and service aging coefficient of 0.92. After linear interpolation to complete the heat treatment cooling rate parameters, a labeled sequence element ID is generated: Sensor-01-T15000. Verification confirms that there are no missing data, and a complete labeled thermal propagation fingerprint feature vector sequence is output, significantly improving the physical interpretability and accuracy of subsequent benchmark template construction.

[0088] S3.2: Based on the cumulative service cycle duration field in the labeled thermal propagation fingerprint feature vector sequence, perform segmented clustering analysis to divide the thermal propagation fingerprint feature vectors of the same physical location in different service time windows into three discretized service stage sets: initial break-in period, stable service period, and aging and decay period, so as to generate a segmented thermal propagation fingerprint feature vector group that characterizes the mold life evolution trajectory, thereby realizing the second-level data conversion from continuous time series to discretized service stage sets.

[0089] The cumulative service life field is read from the labeled thermal propagation fingerprint feature vector sequence and used as the basis for time-dimensional clustering. A fast search and discovery method model based on density peaks is constructed to calculate the local density and relative distance of each sample point. Time window boundary constraints are set for the initial break-in period, stable service period, and aging and decay period. Sample points with service life less than a preset threshold T1 are divided into the initial break-in period set. Sample points with service life between T1 and T2 are divided into the stable service period set. Sample points with service life greater than T2 are divided into the aging and decay period set. Spatial topological recombination is performed on the thermal propagation fingerprint feature vectors in the three sets. A staged thermal propagation fingerprint feature vector group containing stage labels is generated. Through segmented clustering analysis, the continuous time series is transformed into a discretized service stage set, realizing a structured representation of the mold life evolution trajectory.

[0090] For example, the initial break-in period threshold T1 is set to 5000 injection cycles, and the stable service period upper limit T2 is set to 50000 cycles. Thermal propagation fingerprint data of the center position of a mold cavity is collected at 1000, 10000, and 60000 cycles. Data from the 1000-cycle period is categorized into the initial break-in period set because it is less than T1. Data from the 10000-cycle period is categorized into the stable service period set because it falls between T1 and T2. Data from the 60000-cycle period is categorized into the aging and degradation period set because it exceeds T2. The vectors within each set are reorganized, outputting a feature vector group containing stage identifiers, significantly improving the stage-specificity of subsequent baseline template construction.

[0091] S3.3: For each discretized service stage set in the phased thermal propagation fingerprint feature vector group, the mean vector and covariance matrix of the three-dimensional parameters of rise time, peak attenuation rate and phase offset are calculated by weighted statistical normalization method to eliminate single-cycle random noise and extract the typical thermal conduction behavior mode under that stage, thereby generating a phased benchmark fingerprint prototype vector characterizing the standard thermal response of each service stage, completing the third-level data transformation from the original sample set to the phased benchmark fingerprint prototype vector.

[0092] For the phased thermal propagation fingerprint feature vector group, a three-dimensional parameter matrix of rise time, peak attenuation rate and phase offset is extracted for all samples in the same service stage set.

[0093] Outlier detection is performed on the three-dimensional parameter matrix. The Mahalanobis distance method is used to calculate the statistical distance of each sample point relative to the data center at that stage. Abnormal fluctuation data with a distance exceeding 3 times the standard deviation are removed to ensure the purity of the baseline prototype.

[0094] A weighted coefficient allocation model is constructed, which assigns differentiated weights to the retained effective samples based on the stability index of the injection molding cycle and the historical data of sensor signal-to-noise ratio. Samples with high stability cycles are given higher weights to enhance the representativeness of the benchmark.

[0095] Based on a weighted sample set, the weighted arithmetic mean of rise time, peak attenuation rate, and phase offset is calculated to generate a mean vector characterizing the central trend of this service stage. The covariance matrix between parameters of each dimension is calculated to quantify the coupled fluctuation characteristics of parameters such as rise time and phase offset in a specific service stage. The generated mean vector and covariance matrix are encapsulated into a structured data object, defined as the stage-specific benchmark fingerprint prototype vector for this discretized service stage.

[0096] By using weighted statistical normalization, the phased sample set from the previous step is transformed into a phased benchmark fingerprint prototype vector with statistical significance, thereby achieving the expected technical effect of eliminating single-cycle random noise and extracting typical heat conduction behavior patterns.

[0097] For example, during the stable service phase, thermal propagation fingerprint data from 500 injection molding cycles were collected. Samples with a signal-to-noise ratio (SNR) higher than 40 dB were weighted at 1.0, and the rest at 0.5. After filtering out 15 outliers using Mahalanobis distance, the average rise time was calculated to be 12.5 ms, the average peak decay rate was 0.85 °C / s, and the average phase shift was 15.2 degrees. The covariance matrix showed a strong positive correlation between rise time and phase shift. This prototype vector serves as the core benchmark for subsequent dynamic threshold determination, significantly improving the robustness of disturbance identification.

[0098] S3.4: The material thermal property decay model is used to perform cross-stage trend fitting processing on the stage benchmark fingerprint prototype vector. Based on the thermal conductivity decay curve corresponding to the batch of mold material and the residual stress release law determined by the heat treatment state, a continuous interpolation function connecting different discrete service stage sets is constructed to generate a dynamic evolution benchmark fingerprint trajectory curve that can reflect the continuous degradation process of mold performance, realizing the fourth level of data conversion from discrete stage prototype to continuous dynamic evolution benchmark fingerprint trajectory curve.

[0099] The stage-based benchmark fingerprint prototype vector generated by S3.3, as well as the three-dimensional calibration information of mold material batch, heat treatment status, and service cycle, are obtained as input data sources for constructing a continuous evolution trajectory.

[0100] The system calls upon a pre-stored database of material thermal property degradation to retrieve the degradation curve parameters of the thermal conductivity of the current mold steel grade as a function of temperature and time, and extracts the initial thermal conductivity and degradation rate constant.

[0101] Based on the heat treatment process parameter curves, the residual stress distribution field inside the mold is calculated. Combined with the service cycle duration, the influence factor of residual stress release on the local heat conduction path impedance is derived using a stress relaxation model.

[0102] By coupling the thermal conductivity decay curve with the stress release influencing factor, a physical degradation function describing the continuous change of mold thermal conductivity with service time is constructed, and the mapping relationship between fingerprint feature parameters and physical degradation state is established.

[0103] The cubic spline interpolation method is used to smooth the discrete benchmark fingerprint prototype vectors of the initial break-in period, stable service period and aging and degradation period to ensure the continuity of derivatives at the transition points of each stage.

[0104] The interpolation nodes are weighted and corrected based on the physical degradation function to compensate for the nonlinear drift caused by the evolution of the material's microstructure, thereby generating a dynamic evolutionary benchmark fingerprint trajectory curve covering the entire life cycle.

[0105] By using a material thermophysical property degradation model and cross-stage trend fitting, discrete stage-based benchmark fingerprint prototypes are transformed into dynamic evolution benchmark fingerprint trajectory curves that reflect the continuous degradation process of mold performance. This achieves the fourth-level data conversion from discrete stage prototypes to continuous dynamic evolution benchmark fingerprint trajectory curves, providing a high-precision time-varying reference benchmark for subsequent real-time matching.

[0106] For example, for P20 mold steel injection molds, the thermal conductivity decay constant of 0.002 / h and the residual stress release coefficient of 0.85 are extracted. Cubic spline interpolation is performed on the rise time prototype vectors [12ms, 13.5ms, 15.2ms] for the break-in, stabilization, and aging periods. The node weights are corrected using a physical degradation function to generate a continuous trajectory curve. At 5000 hours of service, the predicted rise time of the trajectory curve is 14.1ms, with a deviation of less than 0.3ms from the measured value, significantly improving the accuracy of the benchmark template in characterizing the mold aging state and effectively supporting the accuracy of subsequent difference calculations.

[0107] S3.5: Based on the dynamic evolution benchmark fingerprint trajectory curve and the real-time input current three-dimensional calibration information, perform final matching verification, encapsulate the verified trajectory data into a structured data storage object with version iteration capability, so as to build a benchmark fingerprint template library that supports online query and incremental update operations and dynamically updates with mold aging, and finally complete the fifth level of data transformation from theoretical evolution curve to benchmark fingerprint template library that can be applied in engineering and dynamically updates with mold aging.

[0108] It receives the dynamic evolution baseline fingerprint trajectory curve data generated by S3.4 and the real-time input three-dimensional calibration information of mold material batch, heat treatment status, and cumulative service cycle. The real-time three-dimensional calibration information is standardized and encoded, mapped to query index keys in a high-dimensional feature vector space, ensuring consistency between the input data format and the template library storage structure.

[0109] Based on the query index key value, the theoretical fingerprint coordinate point corresponding to the current service stage is located in the dynamic evolution benchmark fingerprint trajectory curve, and the theoretical predicted values ​​of rise time, peak attenuation rate and phase offset at the coordinate point are extracted.

[0110] Calculate the Euclidean distance between real-time 3D calibration information and historical calibration records, and construct a confidence evaluation function to quantify the degree of matching between the current input data and the known mold state model.

[0111] The matching verification score is calculated using the following formula: in, To calibrate the Euclidean distance of the data in real time, and These are the minimum and maximum distances in the calibration library, respectively.

[0112] When the matching verification score is higher than the preset threshold of 0.85, the verification is deemed successful, and the theoretical prediction value is marked as valid benchmark data; otherwise, the exception handling mechanism is triggered, the benchmark template of the previous period is used, and the deviation log is recorded.

[0113] The verified trajectory data is encapsulated into a structured data storage object in JSON format, which includes a version identifier, timestamp, 3D calibration information, and fingerprint feature vector fields, and is given a unique version number to enable iterative tracking.

[0114] Write the structured object into a NoSQL database that supports concurrent read and write operations, create a composite index based on location tags and version numbers, and build a benchmark fingerprint template library that supports online millisecond-level queries and incremental update operations and is dynamically updated as the mold ages.

[0115] Through the above-mentioned verification, encapsulation, and storage mechanisms, the theoretical evolution curve from the previous step is transformed into a dynamic benchmark template library with engineering usability, enabling the compensation control strategy to accurately adapt to the thermal changes throughout the entire life cycle of the mold.

[0116] like Figure 3As shown, step S4 involves: calculating the difference between the current thermal propagation fingerprint feature vector and the corresponding template in the benchmark fingerprint template library in real time, and logically judging this difference against a preset dynamic threshold that is adaptively adjusted according to the mold's thermal history to generate a region activation command that identifies the local thermal disturbance initiation state. Specifically, this includes: S4.1: Based on the cumulative data of the mold service cycle and the statistical information of the recent thermal cycle count, the historical thermal propagation fingerprint feature vector in the benchmark fingerprint template library is subjected to time-weighted attenuation processing to generate a dynamic benchmark weight coefficient sequence that reflects the current material aging state and thermal fatigue degree of the mold.

[0117] The cumulative service life data and recent thermal cycle count statistics of the mold are read as initial input variables for dynamic weight calculation. A weight allocation model based on the time decay factor is constructed, defining the time-effect coefficient of the current moment relative to the historical benchmark. The time weight of each historical fingerprint sample is calculated using an exponential decay function to match the changes in material sensitivity at different service stages. The recent thermal cycle count is introduced as a correction factor into the weight calculation to nonlinearly compensate for the accelerated aging effect caused by high-frequency thermal shock. The time weight coefficient is applied to the historical thermal propagation fingerprint feature vector sequence in the benchmark fingerprint template library using a weighted summation method. A dynamic benchmark weight coefficient sequence reflecting the current material aging state and thermal fatigue degree of the mold is generated. Through the above processing, static historical fingerprint data is transformed into dynamic weight indicators with time-varying characteristics, realizing the accurate quantification of the thermal evolution law of the mold throughout its entire life cycle, and providing reliable data support for the subsequent generation of adaptive benchmark fingerprints.

[0118] S4.2: The standard thermal propagation fingerprint feature vector stored in the benchmark fingerprint template library is linearly reconstructed using the dynamic benchmark weight coefficient sequence to output an online adaptive benchmark fingerprint feature vector that matches the current physical state of the mold in real time.

[0119] The system receives a dynamic baseline weighting coefficient sequence generated by S4.1, which includes time decay factors corresponding to nodes in each thermally sensitive region of the mold. It then retrieves a standard thermal propagation fingerprint feature vector from the baseline fingerprint template library that matches the current sensor location label. This vector consists of three dimensions: rise time, peak decay rate, and phase offset.

[0120] A linear weighted reconstruction model is constructed, and the dynamic benchmark weight coefficient is used as an adjustment factor to perform element-wise multiplication on each component of the standard thermal propagation fingerprint feature vector.

[0121] Vector normalization is performed to eliminate dimensional bias caused by weight scaling, ensuring that the reconstructed fingerprint vector remains within a physically interpretable numerical range. MinMax normalization is applied to the reconstructed rise time, peak decay rate, and phase offset to map them to the [0,1] interval.

[0122] The normalized parameters are recombined to generate an online adaptive benchmark fingerprint feature vector that matches the current physical state of the mold in real time. This vector reflects the expected thermal response benchmark of the mold during its current service life, eliminating systematic drift caused by material aging.

[0123] By using linear weighted reconstruction and normalization, the dynamic weight coefficients and static standard templates from the previous step are transformed into online adaptive benchmark fingerprint feature vectors that reflect the real-time physical state. This enables the benchmark data to dynamically follow the aging state of the mold, providing a high-confidence comparison reference for subsequent difference calculations.

[0124] S4.3: Perform a multidimensional Euclidean distance metric calculation on the current thermal propagation fingerprint feature vector acquired in real time and the online adaptive reference fingerprint feature vector to quantify the comprehensive deviation between the two in the three dimensions of rise time, peak attenuation rate and phase offset as a real-time difference scalar.

[0125] The system receives the real-time acquired current thermal propagation fingerprint feature vector and the online adaptive baseline fingerprint feature vector generated by S4.2 linear reconstruction, serving as the input data source for a multi-dimensional spatial distance metric. Normalization preprocessing is performed on the three dimensions of rise time, peak decay rate, and phase offset in the two feature vectors to eliminate weight imbalances caused by dimensional differences, ensuring a balanced contribution of each physical parameter in the distance calculation. A three-dimensional Euclidean spatial metric model is constructed, mapping the normalized current fingerprint vector and the baseline fingerprint vector to the same coordinate system. The sum of squared differences between the two in each dimension is calculated to determine the real-time difference scalar. Floating-point precision correction is applied to the calculated difference scalar, retaining four decimal places to match the data precision requirements of the subsequent threshold comparison module. Through the above multi-dimensional Euclidean distance metric calculation, the feature vector from the previous step is transformed into a real-time difference scalar quantifying the degree of local thermal deviation, achieving a highly sensitive numerical characterization of minute thermal disturbances.

[0126] S4.4: Based on the sliding window variance change rate of the real-time difference scalar, combined with the preset basic noise tolerance parameter, an exponential smoothing filtering method is executed to dynamically correct the initial judgment limit, thereby generating an adaptive dynamic threshold parameter that evolves in real time with the thermal history fluctuation of the mold.

[0127] The system receives the real-time difference scalar sequence output from the previous step and constructs a sliding time window of length Q to cover the difference data of the most recent Q injection molding cycles. The variance of the difference scalar within this sliding window is calculated to quantify the dispersion and stability characteristics of recent thermal fluctuations. A preset baseline noise tolerance parameter is introduced, set based on the background noise level of the sensor array under steady-state conditions, serving as the lower limit for threshold adjustment. An exponential smoothing filter method is executed, using a smoothing coefficient to weight and fuse the current variance value with the historical smoothed variance, eliminating transient abrupt interference. The smoothing coefficient is dynamically adjusted according to the mold's thermal history; when a continuous high-temperature service period is detected, the variance weight is increased to enhance sensitivity to slow drift caused by thermal fatigue. The calculated adaptive dynamic threshold parameter is output to the logic judgment module as a dynamic boundary condition for determining whether a local thermal disturbance has been initiated. Through the above processing, the static difference measurement from the previous step is transformed into a dynamic judgment boundary that evolves in real time with the mold's service state, achieving adaptive robustness of the compensation triggering mechanism against mold aging and environmental noise.

[0128] S4.5: Perform a Boolean logic comparison operation between the real-time difference scalar and the adaptive dynamic threshold parameter. When the real-time difference scalar exceeds the adaptive dynamic threshold parameter, immediately trigger an interrupt signal and encapsulate a region activation instruction data packet containing location labels and disturbance intensity information.

[0129] The real-time difference scalar calculated by S4.3 and the adaptive dynamic threshold parameters generated by S4.4 are obtained as the input data source for logical decision-making.

[0130] Boolean logic comparison operations are performed on the real-time difference scalar and the adaptive dynamic threshold parameter to construct a high-speed hardware interrupt triggering mechanism to eliminate the microsecond-level delay caused by software polling.

[0131] When the real-time difference scalar is determined to be significantly greater than the adaptive dynamic threshold parameter, a high-priority interrupt request signal is immediately generated, the current control cycle is frozen, and the mold hot snapshot data is locked.

[0132] Extract the sensor spatial coordinate index corresponding to the moment the interruption is triggered, and combine it with the thermal propagation fingerprint feature vector identifier recorded in step S2 to determine the specific physical location label of the disturbance.

[0133] Based on the ratio of the amplitude of the real-time difference scalar exceeding the threshold, the intensity level of the local thermal disturbance is quantitatively calculated and mapped to a normalized disturbance energy coefficient between 0 and 1.

[0134] The location tag, disturbance energy coefficient, and timestamp are encapsulated into a standardized region activation instruction data packet, which is then directly written into the input buffer of the compensation control strategy module via a shared memory mechanism.

[0135] By using the above Boolean logic comparison and interrupt triggering process, the difference quantification result of the previous step is transformed into a region activation instruction data packet with spatiotemporal positioning information, thereby achieving the expected technical effect of zero-delay response from thermal anomaly perception to compensation decision triggering.

[0136] For example, the adaptive dynamic threshold parameter is set to 0.15, and the real-time collected difference scalar value of a certain hotspot location is 0.22. The system performs a comparison operation, determines that 0.22 > 0.15, and triggers an interrupt. The coordinates of the point (120, 85) mm are extracted, and the disturbance intensity coefficient is calculated as (0.22 - 0.15) / 0.15 = 0.47. An instruction packet containing the coordinates, the intensity of 0.47, and a timestamp is encapsulated and sent to the S5 module. This process takes less than 1 ms, significantly improving the response speed and ensuring that the compensation strategy is loaded before warping deformation occurs.

[0137] Step S5: Based on the region activation instruction, the perturbation propagation mapping relationship pre-stored in the local knowledge graph is invoked to generate a regionalized compensation strategy parameter set for a specific perturbation mode. Specifically, this includes: S5.1: Based on the thermal propagation fingerprint feature vector identifier carried by the region activation instruction, retrieve and extract the disturbance propagation mapping relationship entries that match the identifier in the local knowledge graph to obtain the original association rule dataset containing the spatial influence domain range, expected deformation direction vector and optimal compensation action point coordinates, thereby completing the directional indexing from real-time disturbance signal to historical experience knowledge base.

[0138] S5.2: Perform geometric topological calculation on the spatial influence domain range in the original association rule dataset, and calculate the dynamic weight allocation matrix of the disturbed region by combining the real-time boundary conditions of the local thermal distribution of the current mold, so as to quantify the contribution of each potential compensation point to the suppression of the current thermal disturbance, thereby generating a dynamic weight allocation matrix that characterizes the coupling strength of the region.

[0139] Receive the original association rule dataset output from step S5.1, which includes the spatial influence domain range, the expected deformation direction vector, and the coordinates of the optimal compensation point. Extract the set of vertex coordinates of the spatial influence domain geometric boundaries defined in the original association rule dataset, and construct a local mesh topology based on Delaunay triangulation to discretize and characterize the connectivity of heat conduction paths inside the mold steel body.

[0140] Real-time temperature field data of the local thermal distribution of the mold at the current moment is obtained and mapped onto the nodes of the aforementioned local mesh topology as the initial boundary conditions for heat conduction calculation. For each potential compensation point, the Euclidean distance from it to each mesh node in the spatial influence domain is calculated, and a thermal resistance attenuation model is constructed by combining the thermal conductivity of the mold material.

[0141] A Gaussian kernel function is used to weight the thermal resistance attenuation model, and the thermal influence weight of each potential compensation point on each node in the spatial influence domain is calculated.

[0142] The calculated thermal influence weights are normalized to ensure that the sum of the weights of the same compensation point to all affected nodes is 1, thus eliminating the impact of dimensional differences on subsequent calculations. The normalized weight values ​​are then filled into a pre-defined sparse matrix structure, with row indices corresponding to potential compensation points and column indices corresponding to grid nodes within the spatial influence domain, generating a preliminary dynamic weight allocation matrix.

[0143] A temperature gradient factor based on real-time thermal distribution is introduced to modify the initial dynamic weight allocation matrix. For regions with large temperature gradients, the weight coefficients of the corresponding grid nodes are increased to enhance the suppression of thermal disturbances in those regions. Through matrix multiplication, the modified dynamic weight allocation matrix is ​​linearly combined with the expected deformation direction vector to quantify the overall contribution of each potential compensation point to the suppression of the current thermal disturbance.

[0144] Through the above geometric topology calculation and thermal resistance weighted correction method, the original association rules of the previous step are transformed into a dynamic weight allocation matrix that characterizes the regional coupling strength, realizing the precise quantitative allocation of compensation resources in the spatial dimension, and providing accurate weight basis for the subsequent generation of target temperature field correction curve.

[0145] S5.3: The expected deformation direction vector is projected and transformed using the dynamic weight allocation matrix. Based on the least squares principle, a target temperature field correction curve that can offset the deformation trend is fitted to eliminate the risk of overshoot or under-adjustment of the global compensation strategy in the local area, thereby generating a target temperature field correction curve specifically for the current disturbance mode.

[0146] The system receives the dynamic weight allocation matrix and the expected deformation direction vector generated in the previous steps, which serve as the input data basis for constructing the target temperature field correction curve.

[0147] The expected deformation direction vector is processed by spatial coordinate mapping, which transforms it from the mold geometric coordinate system to the thermodynamic influence domain coordinate system, thus establishing the physical coupling relationship between the deformation trend and the local temperature gradient.

[0148] A temperature field correction optimization model is constructed based on the principle of least squares. The objective function is defined as minimizing the sum of squared residuals between the actual compensated temperature field and the ideal offset temperature field to quantify the fitting accuracy.

[0149] Boundary constraint verification is performed on the target temperature field correction curve to limit the curve peak value to not exceed the thermal fatigue limit threshold of the mold material, and the area outside the curve is forced to zero to maintain the global thermal equilibrium state.

[0150] By using the aforementioned least-squares fitting and boundary constraint processing methods, the dynamic weights and deformation vectors from the previous step are transformed into smooth and physically feasible target temperature field correction curve data, thereby achieving the expected technical effect of eliminating the risk of local overshoot or under-adjustment and generating a dedicated perturbation mode correction curve.

[0151] S5.4: Based on the mapping constraint between the target temperature field correction curve and the coordinates of the optimal compensation point, a multivariable PID parameter optimization method is executed. The preset control gain search space is traversed to calculate the proportional coefficient, integral time, and derivative time combination that minimizes the system response time and overshoot, thereby generating a regional compensation strategy parameter set containing specific numerical settings.

[0152] The system receives the target temperature field correction curve generated by S5.3 and the dynamic weight allocation matrix output by S5.2, using them as boundary constraints and input variables for the multivariable PID parameter optimization method. A three-dimensional control gain search space is constructed, including proportional gain, integral time, and derivative time. The search step size and upper / lower thresholds for each dimension are set based on the response bandwidth of the mold temperature control unit. An improved particle swarm optimization algorithm is used to initialize the population, mapping the position vector of each particle to a set of candidate PID parameter combinations. The sum of squared tracking errors of the target temperature field correction curve is defined as the fitness function. The velocity update of each particle in the search space within the current iteration cycle is calculated, and an inertia weight factor is introduced to balance global exploration and local exploitation capabilities, ensuring rapid convergence in complex nonlinear heat conduction models. A feasibility check is performed on the updated particle positions, eliminating parameter combinations that violate actuator physical limits or cause system instability, retaining the effective solution set that satisfies the stability criterion. The system step response characteristics corresponding to each particle in the effective solution set are evaluated, and rise time and overshoot indices are extracted. A non-dominated solution that balances speed and stability is selected through the Pareto optimal front. The particle position with the highest overall fitness value is selected from the non-dominated solution set, and the optimal combination of proportional coefficient, integral time, and derivative time values ​​is decoded. Through multivariable PID parameter optimization, the target temperature field correction requirement of the previous step is transformed into a regional compensation strategy parameter set containing specific numerical settings, thereby achieving the optimal dynamic compensation control effect for a specific thermal disturbance mode.

[0153] For example, the scaling factor search range is set to 0.5 to 5.0, the integration time search range is 10s to 100s, and the derivative time search range is 0.1s to 5.0s. The initial particle swarm size is 30, the maximum number of iterations is 50, and the inertia weight decreases linearly from 0.9 to 0.4. For a local thermal disturbance scenario, the target temperature field correction curve requires the temperature difference to be controlled within ±0.5℃ within 20s. After iteration, the optimal solution is converged on the 35th iteration, with an output scaling factor of 2.8, an integration time of 45s, and a derivative time of 1.2s. Under this parameter combination, simulation verification shows that the system rise time is 12s and the overshoot is only 3%, which is significantly better than the 18s rise time and 15% overshoot of the traditional empirical tuning method, effectively suppressing the risk of local warping deformation.

[0154] S5.5: Perform validity verification and smoothing filtering on the regional compensation strategy parameter set, eliminate abnormal parameter values ​​that exceed the physical limits of the execution unit and suppress high-frequency noise interference to ensure the stability and executability of the output command, and finally form the final regional compensation strategy parameter set that can directly drive the temperature control unit.

[0155] The system receives an initial regionalized compensation strategy parameter set generated by S5.4, which includes a combination of proportional gain, integral time, and derivative time. Physical boundary constraint checks are performed on each control gain component in the parameter set, and hardware limits such as the maximum allowable drive current, maximum heating power, and minimum cooling response time of the temperature control unit are read. Multidimensional inequality constraints are constructed, and the initial parameters are compared with the hardware limits one by one. For proportional gain or integral time exceeding the upper limit, it is truncated to the corresponding hardware safety threshold to prevent actuator saturation or damage due to excessively strong control signals. A sliding window mid-range filter is applied to the parameter sequence after boundary truncation, with the filter window length set to three control cycles to eliminate parameter spikes caused by instantaneous sensor noise or calculation jitter. A weighted moving average method is used to smooth the filtered parameters, with the weight coefficient decreasing linearly with the time step to ensure that the parameter change rate between adjacent control cycles remains within a preset smooth range. An actuator response dead-zone compensation mechanism is introduced to detect whether parameter values ​​fall within the actuator's nonlinear dead-zone range. If a parameter is detected to fall into the dead zone, a reverse bias is applied based on historical calibration data to offset the control blind zone caused by mechanical hysteresis. Through the above-mentioned validity verification and smoothing filtering, the theoretically optimal parameters generated in the previous step are transformed into a final regional compensation strategy parameter set that conforms to the physical characteristics of the hardware and has high stability, thereby significantly improving the executability of compensation instructions and the robustness of the system.

[0156] Step S6: Based on the regionalized compensation strategy parameter set, load a compensation algorithm sub-model containing only adjustable parameters, and perform short-term response calculation to generate a region-specific compensation control quantity after regional mapping compression. Specifically, this includes: S6.1: Based on the proportional coefficient, integral time, and derivative time combination defined in the parameter set of the regionalized compensation strategy, the pre-stored general full injection molding compensation model is subjected to structural pruning and parameter freezing. Redundant network nodes and fixed weight layers that are irrelevant to the current thermal disturbance influence domain are removed to construct a compensation algorithm sub-model instance that retains only the adjustable weights for specific disturbance modes. This completes the first-level model reconstruction transformation from the general full injection molding compensation model to the compensation algorithm sub-model instance.

[0157] The proportional coefficient, integral time, and derivative time combination in the parameter set of the regional compensation strategy are received as control constraints for model reconstruction.

[0158] Read the pre-stored general full injection molding compensation model, which includes a multi-layer neural network structure and a weight connection matrix for global mold nodes.

[0159] Based on the spatial influence domain range generated in step S5, a binary mask vector is constructed to mark network nodes located outside the influence domain as redundant nodes.

[0160] Perform structural pruning on network nodes marked as redundant, severing all weight connections between them and adjacent layers, and freezing the corresponding weight parameters to zero.

[0161] Network nodes and their connection paths that are strongly coupled to the thermal disturbance influence domain are retained to form a sparse subnetwork topology.

[0162] By mapping the scaling factor in the parameter set of the regional compensation strategy to the gain adjustment factor of the sub-network output layer, linear scaling of the control strength is achieved.

[0163] The integral time parameter is converted into the state memory decay coefficient of the hidden layer of the sub-network to suppress the excessive accumulation effect of historical error.

[0164] The differential time parameter is mapped to the width of the differential filter window in the input layer, thereby enhancing the sensitivity to the rate of change of thermal state.

[0165] The pruned subnetwork is subjected to connectivity verification to ensure that there is a complete and effective signal propagation path from the input layer to the output layer.

[0166] The initial values ​​of the non-tunable weights in the fixed subnetwork are set, and only the weights of the terminal neurons corresponding to the optimal compensation action point are made trainable.

[0167] By using structural pruning and parameter freezing, the parameter set of the regional compensation strategy from the previous step is transformed into a sub-model instance of the compensation algorithm, thereby achieving the expected technical effects of reducing computational load and improving response speed.

[0168] For example, the general full-volume injection molding compensation model includes 128 input nodes, 3 hidden layers (64 nodes per layer), and 32 output nodes. The spatial influence domain covers 10% of the local area of ​​the mold, corresponding to output node indices 5-8. A mask vector is constructed to prune output nodes with indices other than 5-8 and their upstream redundant connections, reducing the number of nodes retained to 15% of the original model. The scaling factor of 1.2 is mapped to an output layer gain factor of 1.2; the integration time of 0.5s is converted to a hidden layer attenuation coefficient of 0.95; and the derivative time of 0.1s sets the input difference window to 3 sampling periods. Only the last-level weights (256 parameters) corresponding to output nodes 5-8 are made adjustable, while the remaining 12,000+ parameters are frozen. The sub-model inference time is reduced from 50ms to 2ms, meeting the requirements of sub-periodic dynamic compensation.

[0169] S6.2: Using the compensation algorithm sub-model instance, the target temperature field correction curve and dynamic weight allocation matrix generated in the previous step are received as real-time input tensors. Forward propagation inference is performed to solve the nonlinear mapping relationship between local thermal deviation and actuator drive signal, thereby generating the original regional compensation drive vector characterizing the theoretical compensation intensity, realizing the second-level data deduction and transformation from multi-dimensional control parameter input to the original regional compensation drive vector.

[0170] The system receives a sub-model instance of the compensation algorithm, which has been loaded with proportional, integral, and differential parameter weights optimized for the current thermal disturbance mode. The target temperature field correction curve generated in the previous steps is discretized into a time-series vector, serving as the model's state expectation input. The dynamic weight allocation matrix is ​​flattened into a spatial dimension vector, serving as the model's coupling strength constraint input. A forward propagation path for a nonlinear mapping network is constructed, where the hidden layer uses the ReLU activation function to capture the nonlinear hysteresis characteristics of heat conduction. The net input values ​​of the hidden layer neurons are calculated through weighted summation. Activation function processing is applied to the net input values ​​to generate the hidden layer output vector. The hidden layer output vector is passed to the output layer, where a linear transformation is performed to solve for the actuator drive signal. The output layer nodes correspond to the pulse width modulation duty cycle and phase offset of each temperature control unit. A regularization term is introduced to suppress overfitting and ensure the model's generalization ability in short-term responses. The final drive values ​​of the output layer are calculated, forming the original region compensation drive vector. This vector contains the theoretical control strength of each activated execution unit.

[0171] By using forward inference operations of a lightweight sub-model, the target temperature field correction requirements are transformed into specific actuator drive commands, enabling rapid derivation from multi-dimensional control parameters to the original region compensation drive vector, significantly reducing computational latency and improving real-time response.

[0172] S6.3: Perform spatial mask filtering processing based on the mold actuator topology mapping table on the original region compensation driving vector, force the component in the vector corresponding to the non-strongly coupled temperature control unit to zero and retain the value of the effective action channel, so as to generate a sparse region compensation driving matrix that strictly matches the thermal disturbance influence domain only in the spatial dimension, and complete the third-level spatial compression transformation from continuous global driving vector to sparse region compensation driving matrix.

[0173] The system receives the original region compensation drive vector generated in the previous steps. This vector contains the theoretical control signal components of all temperature control units in the mold. A pre-stored mold actuator topology mapping table is loaded, which defines the spatial coupling relationship matrix between the physical coordinates of each actuator and the thermal disturbance propagation path. The original region compensation drive vector is analyzed for dimensionality, decomposing it into effective channel components strongly coupled to the spatial influence domain and redundant channel components in irrelevant regions. Based on the coupling strength threshold in the topology mapping table, a binary spatial mask matrix is ​​constructed, where strongly coupled positions are marked as 1 and weakly coupled positions as 0. An element-wise Hadamard product operation is performed between the original region compensation drive vector and the binary spatial mask matrix. Components with a mask coefficient of 0 in the result are forcibly zeroed to eliminate invalid intervention of the global control strategy in non-target regions. The values ​​of components with a mask coefficient of 1 are retained, forming a sparsed region compensation drive matrix that strictly matches the thermal disturbance influence domain only in the spatial dimension. By using spatial masking filtering, the continuous global driving vector from the previous step is transformed into sparse region compensation driving matrix data, achieving the expected technical effect of precise spatial focusing of actuator actions and significant reduction in computational load.

[0174] For example, for a thin-walled toy mold, the original region compensation drive vector is 64-dimensional, corresponding to 64 temperature control points. The topology mapping table identifies that the thermal disturbance influence domain only covers the three execution units numbered 12, 13, and 28, whose coupling strength is greater than 0.85. A 64-dimensional binary mask matrix is ​​constructed, with only bits 12, 13, and 28 set to 1, and the rest set to 0. After performing the Hadamard product operation, except for bits 12, 13, and 28 which retain the pulse width modulation duty cycle parameter calculated from the original PID, the remaining 61 bits are all set to zero. The proportion of non-zero elements in the generated sparse region compensation drive matrix is ​​only 4.6%, which significantly reduces the computational load of subsequent timing correction and ensures that only local microchannel solenoid valves are activated, avoiding unnecessary fluctuations in the temperature field of the entire mold.

[0175] S6.4: Based on the pulse width modulation duty cycle parameter and phase offset parameter contained in the sparse region compensation driving matrix, and combined with the physical response hysteresis characteristic curves of the injection molding machine heating rod and cooling valve, a time series prediction correction method is executed to compensate for the dynamic delay error of the actuator, thereby generating a precise region compensation control quantity sequence after time-series pre-compensation correction, realizing the fourth-level time domain correction conversion from the static spatial driving matrix to the precise region compensation control quantity sequence.

[0176] Receive the sparse region compensation driving matrix, extract the pulse width modulation duty cycle parameter and phase offset parameter corresponding to each active execution channel, and use them as the initial control variables for timing pre-compensation calculation.

[0177] By calling the pre-stored thermal inertia response curve of the injection molding machine heating rod and the fluid dynamic delay curve of the cooling valve, a dynamic characteristic model library of the actuator, including the heating time constant, cooling lag time and valve opening dead zone, is constructed.

[0178] Based on the current temperature field distribution of the mold steel, the matching physical response hysteresis parameters are retrieved from the model library to determine the time delay from receiving the command to generating an effective heat exchange action for each temperature control unit.

[0179] The discrete-time predictive control method is adopted to reverse the target temperature field correction curve on the time axis. The amount of the shift is equal to the calculated time delay, so as to generate the expected control trajectory ahead of the current moment.

[0180] The calculated control sequence is subjected to saturation limiting processing to ensure that the pulse width modulation duty cycle is within the physical allowable range of the actuator, and to prevent the actuator from being overloaded due to excessive advance compensation.

[0181] By using time series prediction and correction processing, the static spatial driving matrix of the previous step is transformed into a precise regional compensation control quantity sequence with time-series foresight, so as to achieve precise synchronization between compensation action and thermal disturbance propagation in the time dimension and eliminate control error caused by execution lag.

[0182] S6.5: The precise area compensation control quantity sequence is encapsulated into a standard data frame conforming to the industrial fieldbus communication protocol, and the current heat propagation fingerprint feature vector identifier is attached as a check code to generate an area-specific compensation control quantity data packet that can be directly sent to the underlying drive circuit after area mapping compression, thus completing the fifth-level protocol encapsulation conversion from internally calculated variables to area-specific compensation control quantity data packets after area mapping compression.

[0183] Step S7: The temperature control unit, strongly coupled with the thermal disturbance influence domain, is driven to operate according to the region-specific compensation control quantity, while keeping other unrelated actuators in a silent state, thus completing the coordinated adjustment of the local thermal distribution of the mold. Specifically, this includes: S7.1: Based on the spatial influence domain coordinate information contained in the region-specific compensation control quantity, perform index query processing on the pre-stored mold actuator topology mapping table to filter out a set of candidate temperature control units that have a strong coupling relationship with the spatial influence domain, and generate an actuator activation list containing a unique device identifier.

[0184] Receive the region-specific compensation control data packet output from step S6, which has been compressed by region mapping, and parse the spatial influence domain coordinate set and disturbance intensity identifier contained therein. Call the mold actuator topology mapping table pre-stored in local memory, which records the physical installation coordinates, thermal conduction coupling coefficient matrix, and unique device identifier of all temperature control units.

[0185] Based on the set of coordinates of the spatial influence domain, all candidate temperature control units located within this geometric range are retrieved from the topology mapping table. The Euclidean distance between the center coordinates of each candidate execution unit and the centroid of the spatial influence domain is calculated, and a subset of execution units whose distance is less than a preset coupling radius threshold is selected.

[0186] For the selected subset of execution units, their corresponding thermal conduction coupling coefficients are extracted. A weighted sorting method is used to prioritize the execution units according to their coupling coefficients from high to low, ensuring that strongly coupled units are selected first to maximize heat exchange efficiency.

[0187] Set an upper limit on the number of activations, and select the top-ranked execution units to form the final target actuator set. Read the unique device identifier of each unit in the target actuator set from the topology mapping table, and assemble the data list according to a predetermined communication protocol format.

[0188] By using index lookup and coupling degree sorting, the regional compensation control quantity from the previous step is transformed into an actuator activation list containing unique device identifiers, achieving the expected technical effect of precise local thermal coordinated regulation without increasing the number of hardware components.

[0189] S7.2: Based on the unique device identifier in the actuator activation list, send a forced silence command to the remaining unrelated temperature control units not listed in the list to lock their current output power and block any interference from global control signals, thereby generating an actuator state isolation environment across the entire mold range.

[0190] S7.3: Using the pulse width modulation duty cycle parameter and phase offset parameter defined in the region-specific compensation control quantity, the drive signal reconstruction processing is performed on the candidate temperature control unit in the actuator activation list to generate a microchannel solenoid valve opening and closing sequence or a pulsed Peltier module current waveform with specific timing characteristics.

[0191] The region-specific compensation control data package is parsed, and the pulse width modulation duty cycle parameters and phase offset parameters of each candidate temperature control unit in the actuator activation list are extracted to construct the basic instruction set for drive signal reconstruction.

[0192] For microchannel solenoid valve actuators, the high-level duration within a single control cycle is calculated based on the extracted duty cycle parameters. The turn-on delay time relative to the global clock reference is determined by combining the phase offset parameters, thereby generating a discretized switching timing logic sequence.

[0193] For pulsed Peltier module-type execution units, the duty cycle parameter is mapped to the equivalent average current amplitude, the phase offset parameter is used to adjust the starting firing angle of the current waveform, and a continuous current drive waveform with a specific frequency and phase is synthesized by sinusoidal pulse width modulation.

[0194] A time-slice polling mechanism is used to perform spatiotemporal alignment processing on the mixed-type actuator drive signals to ensure that temperature control units with different physical characteristics generate a synergistic heat flow exchange effect on the local thermal disturbance propagation path of the mold, thereby eliminating the compensation cancellation effect caused by the difference in response speed.

[0195] The generated drive signal sequence is subjected to dead time insertion and boundary limiting to prevent electromagnetic interference or mechanical wear of the actuator caused by excessively rapid signal transitions. The output is a microchannel solenoid valve opening and closing sequence or a pulsed Peltier module current waveform that has been verified for integrity.

[0196] By using the above signal reconstruction processing method, the regional compensation strategy parameters of the previous step are transformed into underlying hardware driving instructions with precise timing characteristics, so as to achieve precise synchronization between compensation action and thermal disturbance propagation in time and space, and significantly improve the response speed and suppression effect of local thermal regulation.

[0197] For example, for the microchannel solenoid valve EV-03, the duty cycle is extracted to be 45% and the phase offset to be 12ms. The control cycle is set to 100ms, and the high-level duration is calculated to be 45ms, with the on-time set at 12ms. A timing sequence is generated, with the on state from 12ms to 57ms and the off state for the rest of the time. For the Peltier module PT-07, the duty cycle is extracted to be 60% and the phase offset to be 5ms. A maximum current of 5A is mapped to obtain a target average current of 3A. Based on a 1kHz carrier frequency, a sinusoidal current waveform with an initial phase lag of 5ms and amplitude modulation is generated. The two are time-aligned to ensure simultaneous activation 15ms before the thermal disturbance reaches its peak. A drive data packet containing the EV-03 switching sequence and the PT-07 current waveform is output, verified, and then sent out, effectively eliminating local hot spots and significantly reducing temperature fluctuations in that area.

[0198] S7.4: The microchannel solenoid valve opening and closing sequence or the pulsed Peltier module current waveform is sent to the corresponding physical interface circuit for power amplification processing to drive the selected temperature control unit to generate directional heat flow exchange action, thereby forming a reverse temperature gradient field against local thermal disturbances.

[0199] S7.5: Monitor the real-time propagation effect of the reverse temperature gradient field in the mold steel medium, and compare and verify the actual thermal response data with the expected deformation direction to confirm that the goal of coordinated adjustment of the local thermal distribution of the mold is achieved without increasing the number of hardware, and complete the closed-loop execution record of this regional compensation response.

[0200] Receive real-time monitoring data of the local thermal distribution of the mold under the action of the reverse temperature gradient field generated in step S7.4. This data includes the transient voltage signal sequence collected by the pyroelectric sensor array embedded in the key nodes of the mold.

[0201] The transient voltage signal sequence is processed by timestamp alignment and spatial topology mapping. The physical coordinates of each sensor node are associated with the thermal response data it collects, and a two-dimensional heat dissipation force map matrix reflecting the temperature field distribution on the mold surface at the current moment is constructed.

[0202] Based on the two-dimensional heat dissipation force map matrix, the temperature gradient vector of the center point of the affected area and its adjacent buffer zone is extracted, and the residual vector between the actual temperature change rate and the expected compensation target temperature change rate is calculated.

[0203] The deviation between the actual thermal response trajectory and the expected deformation suppression trajectory in the multidimensional feature space is calculated using the Euclidean distance metric method, and the effect of the reverse temperature gradient field on the local thermal disturbance is quantitatively evaluated.

[0204] When the calculated deviation scalar is less than the preset convergence threshold, it is determined that the goal of coordinated adjustment of local thermal distribution has been achieved, and a compensation success flag is generated.

[0205] The thermal propagation fingerprint feature vector, region-specific compensation control parameters, and final deviation scalar of this compensation process are packaged and written into the historical execution record database to form a closed-loop execution log.

[0206] Through the above-mentioned real-time monitoring and comparison verification process, the effect of the reverse temperature gradient field in the previous step is transformed into a quantitative compensation effectiveness index, realizing precise local thermal control and closed-loop traceability of the mold without increasing the number of hardware components.

[0207] For example, for a thin-walled toy shell injection mold, five key monitoring nodes were selected. The convergence threshold was set to 0.5℃. In actual operation, the sensor collected node temperatures of [85.2, 84.8, 85.5, 84.9, 85.1]℃, and the expected compensation temperature was [85.0, 85.0, 85.0, 85.0, 85.0]℃. Substituting these values ​​into the formula, the deviation D was calculated to be 0.38℃, which is less than the threshold of 0.5℃. The system determined that the compensation was effective, recorded the fingerprint vector [12.5ms, 0.8% / s, 15°] and the control quantity PWM duty cycle of 65%, and stored them in the database for subsequent model optimization, significantly improving the confidence of single compensation.

[0208] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0209] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0210] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A toy injection molding compensation and control method for simulated warpage deformation of thin-walled parts, characterized in that, Specifically, it includes: S1: Deploy a high time-resolution micro pyroelectric sensor array in the key thermally sensitive area of ​​the mold to acquire the original thermal response signal sequence containing multiple tags at different locations; S2: Analyze the original thermal response signal sequence to obtain the thermal propagation fingerprint feature vector; S3: Construct a benchmark fingerprint template library based on thermal propagation fingerprint feature vectors; S4: Calculate the difference between the current thermal propagation fingerprint feature vector and the corresponding template in the benchmark fingerprint template library in real time, and make a logical judgment between the difference and the preset dynamic threshold that is adaptively adjusted according to the thermal history of the mold to generate a region activation command that identifies the local thermal disturbance start state. S5: According to the region activation instruction, invoke the perturbation propagation mapping relationship pre-stored in the local knowledge graph to generate a regionalized compensation strategy parameter set for a specific perturbation mode; S6: Based on the regional compensation strategy parameter set, load the compensation algorithm sub-model containing only adjustable parameters, and perform short-term response calculation to generate the region-specific compensation control quantity after regional mapping compression; S7: Drive the temperature control unit that is strongly coupled with the thermal disturbance influence domain according to the region-specific compensation control quantity, while keeping the other unrelated actuators in a silent state, to complete the coordinated adjustment of the local thermal distribution of the mold.

2. The toy injection molding compensation control method for warpage deformation simulation of thin-walled parts according to claim 1, characterized in that, The process of analyzing the original thermal response signal sequence to obtain the thermal propagation fingerprint feature vector includes: The original thermal response signal sequence is subjected to time-frequency joint analysis and processing to extract three dimensions of features: rise time, peak attenuation rate, and phase offset relative to neighboring sensors, thereby generating a thermal propagation fingerprint feature vector characterizing the local thermal dynamics.

3. The toy injection molding compensation control method for warpage deformation simulation of thin-walled parts according to claim 1, characterized in that, The construction of the benchmark fingerprint template library based on thermal propagation fingerprint feature vectors includes: Based on the three-dimensional calibration information of mold material batch, heat treatment status and service cycle, the thermal propagation fingerprint feature vectors of the same physical location under different service stages are normalized and compared to construct a benchmark fingerprint template library.

4. The toy injection molding compensation control method for simulated warpage deformation of thin-walled parts according to claim 1, characterized in that, The S1 specifically includes: A location tag mapping table for heat-sensitive areas was obtained based on a miniature pyroelectric sensor array; A multi-point distributed thermal sensing hardware network is constructed based on a thermally sensitive area location label mapping table. A multi-point distributed thermal state sensing hardware network is used to collect data and process it to obtain a discrete original thermal response signal sequence. The discretized original thermal response signal sequence is processed using a sliding window denoising and baseline drift correction method to remove noise components introduced by environmental electromagnetic interference and the sensor's own zero-point drift, and outputs a preprocessed, high signal-to-noise ratio standardized original thermal response signal sequence.

5. The toy injection molding compensation control method for simulated warpage deformation of thin-walled parts according to claim 4, characterized in that, The step of obtaining the location tag mapping table of the heat-sensitive area based on the micro pyroelectric sensor array includes: Based on the distribution information of the main path nodes for thermal disturbance propagation identified by the three-dimensional thermal conduction simulation model of the mold, the thermally sensitive area of ​​the mold steel structure is divided, the optimal spatial layout coordinate set of the micro pyroelectric sensor array on the mold surface is determined, and the location label mapping table of the thermally sensitive area is obtained according to the optimal spatial layout coordinate set.

6. The toy injection molding compensation control method for warpage deformation simulation of thin-walled parts according to claim 4, characterized in that, The construction of a multi-point distributed thermal sensing hardware network based on a thermally sensitive area location label mapping table includes: Based on the location tag mapping table of the heat-sensitive area, a high time resolution micro pyroelectric sensor array is installed at the determined optimal spatial layout coordinates, and the sampling clock synchronization mechanism of each sensor is configured to set the sampling interval parameter, thereby constructing a multi-point distributed thermal sensing hardware network with a unified time base.

7. The toy injection molding compensation control method for simulated warpage deformation of thin-walled parts according to claim 4, characterized in that, The process of acquiring data using a multi-point distributed thermal sensing hardware network and processing it to obtain a discrete original thermal response signal sequence includes: The multi-point distributed thermal sensing hardware network is used to collect transient temperature change data in real time during the injection molding process. The analog voltage signal output by each sensor is converted from analog to digital and timestamped to generate a discrete original thermal response signal sequence with a unique location tag and strict time synchronization.

8. The toy injection molding compensation control method for simulated warpage deformation of thin-walled parts according to claim 5, characterized in that, The three-dimensional heat conduction simulation model of the mold uses the K-means clustering algorithm to divide the heat-sensitive sub-regions. Combined with the cooling water channel layout constraints, the sensor placement points with the maximum response gain and optimal spatial coverage are selected through sensitivity analysis.

9. The toy injection molding compensation control method for warpage deformation simulation of thin-walled parts according to claim 2, characterized in that, The phase offset is calculated by cross-correlation of the thermal response signals of adjacent sensor nodes, and the number of sampling points and angle difference are converted by the lag time. The three-dimensional vector of rise time, peak attenuation rate and phase offset is comprehensively used to characterize the local heat propagation features.

10. The toy injection molding compensation control method for simulated warpage deformation of thin-walled parts according to claim 1, characterized in that, The baseline fingerprint template library contains material batch numbers, heat treatment parameter curves, and service cycle data. It is divided into three stages: initial break-in period, stable service period, and aging and degradation period. The mean and covariance of fingerprints are collected for each stage, and weighted statistical normalization is used to eliminate random noise and outliers.

Images