A method and system for optimizing printhead temperature control

Abstract

The application discloses a kind of print head temperature control optimization method and system, method includes: real-time acquisition print head temperature data, by fast Fourier transform temperature fluctuation is decomposed into low-frequency steady component and high-frequency transient component;According to print material type and ambient temperature, predict material thermal response characteristic curve, generate thermal response compensation coefficient;Low-frequency steady component is generated basic heating power using PID controller, while extracting amplitude feature vector from high-frequency transient component, the amplitude feature vector and the thermal response compensation coefficient are input into pre-trained pulse neural network SNN, and output dynamic compensation power;Based on basic heating power and dynamic compensation power, generate final heating instruction for print head, and constrain instruction output range by anti-saturation limiter, to realize the dynamic control optimization of print head temperature.Utilize the embodiment of the application, the precise dynamic regulation and control of print head temperature can be realized, and the influence of temperature fluctuation on print quality is reduced.

Description

Technical Field

[0001] This invention belongs to the field of printing technology, specifically a method and system for optimizing printhead temperature control. Background Technology

[0002] Existing 3D printhead temperature control methods mostly employ a single PID control strategy, which struggles to handle the complex thermal dynamics during printing. Printhead temperature is susceptible to material properties, ambient temperature fluctuations, and high-frequency thermal interference, leading to insufficient temperature control accuracy and consequently affecting print quality. Traditional methods lack decomposition processing of the frequency domain characteristics of the temperature signal and do not fully consider the nonlinear characteristics of the material's thermal response, resulting in response lag or overshoot. Summary of the Invention

[0003] The purpose of this invention is to provide a printhead temperature control optimization method and system to overcome the shortcomings of the prior art, achieve precise dynamic control of printhead temperature, and reduce the impact of temperature fluctuations on print quality.

[0004] One embodiment of this application provides a printhead temperature control optimization method, the method comprising: Real-time acquisition of printhead temperature data; temperature fluctuations are decomposed into low-frequency steady-state components and high-frequency transient components using fast Fourier transform. Based on the type of printing material and the ambient temperature, the thermal response characteristic curve of the material is predicted, and a thermal response compensation coefficient is generated. The prediction is based on a mapping relationship library between the material's specific heat capacity, thermal conductivity, and ambient temperature. A PID controller is used to generate the basic heating power for the low-frequency steady-state component, while an amplitude feature vector is extracted from the high-frequency transient component. The amplitude feature vector and the thermal response compensation coefficient are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. Based on the base heating power and the dynamic compensation power, a final heating command for the printhead is generated, and the command output range is constrained by an anti-saturation limiter to achieve dynamic control and optimization of the printhead temperature.

[0005] Optionally, the real-time acquisition of printhead temperature data, through Fast Fourier Transform, decomposes temperature fluctuations into low-frequency steady-state components and high-frequency transient components, including: Raw temperature data stream is generated by acquiring multiple points on the surface of the printhead using a high-precision thermocouple array at a sampling rate of 1kHz. The raw temperature data stream is input into an adaptive sliding window filter, and the window size is dynamically adjusted according to the noise statistics to output a noise-reduced temperature sequence. The noise-reduced temperature sequence is processed by Fast Fourier Transform, and the Hanning window function is dynamically selected for spectrum analysis to output a full-band temperature spectrum. Based on the full-band temperature spectrum, an energy accumulation algorithm is applied to automatically determine the 0.5Hz segmentation threshold and output the separated low-frequency steady-state component and high-frequency transient component.

[0006] Optionally, the step of predicting the thermal response characteristic curve of the material based on the printing material type and ambient temperature, and generating a thermal response compensation coefficient, wherein the prediction is based on a mapping relationship library of material specific heat capacity, thermal conductivity, and ambient temperature, including: The material identification sensor obtains the printing material type code, and at the same time reads the ambient temperature sensor data, outputting the material type identifier and the real-time ambient temperature value. Input the material type identifier into the material thermal property database, retrieve the corresponding specific heat capacity and thermal conductivity benchmark values, and output the material basic thermal property vector. Input the material's basic thermal property vector and the ambient temperature value into the thermal response prediction model, calculate the temperature compensation offset, and output the preliminary thermal response curve. The initial thermal response curve is input into the real-time calibration module, and the curve parameters are corrected by combining historical temperature fluctuation data to output an optimized thermal response curve. The time constant and gain factor are extracted from the optimized thermal response curve, and the normalization coefficient is calculated by weighted fusion to output the thermal response compensation coefficient.

[0007] Optionally, the step of using a PID controller to generate the base heating power for the low-frequency steady-state component, while extracting the amplitude feature vector from the high-frequency transient component, and inputting the amplitude feature vector and the thermal response compensation coefficient into a pre-trained spiking neural network (SNN) to output dynamic compensation power includes: The low-frequency steady-state component is input into the adaptive PID controller, and the control parameters are automatically tuned according to the temperature change rate, and the basic heating power signal is output. An envelope detection algorithm is applied to the high-frequency transient components to extract peak value, root mean square and zero-crossing rate features, and output amplitude feature vector. The amplitude eigenvector and the thermal response compensation coefficient are concatenated to generate an enhanced eigenvector. The enhanced feature vector is input into a pre-trained spiking neural network model, and the original compensation power is output through spatiotemporal event-driven processing. The original compensation power is input into the exponential smoothing filter, and the smoothing factor is adjusted according to the frequency of temperature fluctuations to output dynamic compensation power.

[0008] Optionally, the step of generating a final heating command for the printhead based on the base heating power and the dynamic compensation power, and constraining the command output range through an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature, includes: The basic heating power and dynamic compensation power are input into the dynamic weighted fusion unit, the weight ratio is adjusted according to the temperature deviation, and the initial heating command is output. The initial heating command is input into the anti-saturation limiter, and the output range is limited by the dual threshold constraint algorithm to output a safe heating command. The safe heating command is input into the thermal inertia compensation module, and phase correction is performed in combination with the printhead quality parameters to output an optimized heating command. The optimized heating command is converted into a PWM drive signal and output to the printhead heating element to perform temperature control.

[0009] Another embodiment of this application provides a printhead temperature control optimization system, the system comprising: The acquisition module is used to acquire printhead temperature data in real time and decompose temperature fluctuations into low-frequency steady-state components and high-frequency transient components through fast Fourier transform. The prediction module is used to predict the thermal response characteristic curve of the material based on the type of printing material and the ambient temperature, and generate a thermal response compensation coefficient. The prediction is based on a mapping relationship library between the material's specific heat capacity, thermal conductivity and ambient temperature. The extraction module is used to generate a basic heating power from the low-frequency steady-state component using a PID controller, and simultaneously extract the amplitude feature vector from the high-frequency transient component. The amplitude feature vector and the thermal response compensation coefficient are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. The generation module is used to generate a final heating command for the printhead based on the base heating power and the dynamic compensation power, and to constrain the command output range through an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature.

[0010] Another embodiment of this application provides a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any of the preceding claims when running.

[0011] Another embodiment of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any of the preceding claims.

[0012] Compared with existing technologies, the present invention provides a printhead temperature control optimization method that collects printhead temperature data in real time and decomposes temperature fluctuations into low-frequency steady-state components and high-frequency transient components using a fast Fourier transform. Based on the printing material type and ambient temperature, the method predicts the material's thermal response characteristic curve and generates a thermal response compensation coefficient. A PID controller is used to generate a base heating power for the low-frequency steady-state component, while an amplitude feature vector is extracted from the high-frequency transient component. This amplitude feature vector, along with the thermal response compensation coefficient, is input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. Based on the base heating power and the dynamic compensation power, a final heating command for the printhead is generated, and the command output range is constrained by an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature. This enables precise dynamic regulation of the printhead temperature and reduces the impact of temperature fluctuations on print quality. Attached Figure Description

[0013] Figure 1 A hardware structure block diagram of a computer terminal for a printhead temperature control optimization method provided in an embodiment of the present invention; Figure 2 A schematic flowchart illustrating a printhead temperature control optimization method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a printhead temperature control optimization system provided in an embodiment of the present invention. Detailed Implementation

[0014] 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.

[0015] This invention first provides a method for optimizing printhead temperature control, which can be applied to electronic devices, such as computer terminals, specifically ordinary computers.

[0016] The following detailed explanation uses a computer terminal as an example. Figure 1 This is a hardware structure block diagram of a computer terminal for a printhead temperature control optimization method provided in an embodiment of the present invention. (See diagram below.) Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0017] Non-volatile storage media can store an operating system and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any printhead temperature control optimization method.

[0018] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0019] The internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any printhead temperature control optimization method.

[0020] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0021] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0022] See Figure 2 The present invention provides a method for optimizing printhead temperature control, which may include the following steps: S201 collects printhead temperature data in real time and decomposes temperature fluctuations into low-frequency steady-state components and high-frequency transient components through fast Fourier transform. Specifically, a high-precision thermocouple array can be used to collect raw temperature data at multiple points on the surface of the printhead at a sampling rate of 1kHz, generating a raw temperature data stream. Deployment and data acquisition of thermocouple arrays The printhead surface integrates a high-precision thermocouple array, which consists of multiple miniature K-type thermocouples (temperature range -200°C to 1250°C) arranged in a grid pattern, covering key locations such as the nozzle area, the core area of ​​the heating block, and the heat dissipation boundary. Each thermocouple is isolated from the printhead metal substrate by a magnesium oxide insulation layer, ensuring the independence of the temperature measurement point. During printing, the array synchronously acquires the voltage signal (in millivolts mV) of all temperature measurement points at a sampling rate of 1kHz (i.e., 1000 times per second). This high-frequency sampling capability can capture instantaneous temperature fluctuations (such as the quenching effect when the nozzle extrudes material). The thermocouple signals are transmitted to the signal conditioning module via an electromagnetic interference (EMI) shielded cable, where a low-temperature drift instrumentation amplifier (such as Analog Devices' AD8421) amplifies the weak thermoelectric potential (typically 40μV / °C) to a processable range (0-5V) while suppressing common-mode noise (CMN).

[0023] Generation and synchronization of raw data streams The amplified analog signal is input to a multi-channel synchronous sampling ADC (Analog-to-Digital Converter), such as TI's ADS131M08 (24-bit resolution, 8-channel synchronous sampling). The ADC uses strict timing control (clock jitter <1ns) to ensure all channels complete synchronous digitization within a 1-millisecond (ms) time window, avoiding temperature field distortion caused by phase differences between channels. The digitized temperature values ​​(in degrees Celsius °C) are packaged into data packets (DPs) by timestamp (TS) and channel ID (CID). Each DP contains: a timestamp (accurate to microseconds μs), a channel ID (identifying the temperature measurement point location), and a temperature value (floating-point format). The data packets are transmitted to the main control unit via a high-speed SPI (Serial Peripheral Interface) bus and buffered in chronological order as a raw temperature data stream (RTDS). The data stream is stored in the form of a ring buffer, which can hold 10 seconds of data (i.e., 10,000 sampling points) and supports real-time read and write operations.

[0024] Environmental interference suppression and data integrity assurance To address the issue of thermocouple contact resistance variations caused by printing platform vibration, the system employs constant current excitation (CCE, typically 1mA) instead of traditional voltage excitation, significantly reducing the impact of contact resistance fluctuations on temperature measurement accuracy. Simultaneously, a self-calibration circuit is embedded in the signal conditioning stage, automatically performing cold junction compensation (CJC) calibration at midnight daily: reading data from an ambient temperature sensor (such as the DS18B20) and dynamically correcting reference junction (cold junction) temperature drift using a Type K lookup table. Furthermore, the data stream transmission process utilizes a CRC-16 (Cyclic Redundancy Check) algorithm to detect transmission errors, and abnormal data packets trigger an immediate retransmission mechanism to ensure the integrity and timeliness of the original data stream.

[0025] The raw temperature data stream is input into an adaptive sliding window filter, and the window size is dynamically adjusted according to the noise statistics to output a noise-reduced temperature sequence. Real-time analysis of noise characteristics The Adaptive Sliding Window Filter (ASWF) first performs noise statistical analysis on the raw temperature data stream (RTDS). During the initialization phase (e.g., 5 seconds after the print job starts), the filter truncates a segment of data with a fixed window size (e.g., 50 milliseconds, or 50 sampling points) and calculates its noise statistical features (NSF), including: Standard Deviation (SD): Measures the magnitude of noise fluctuation; Zero-Crossing Rate of Autocorrelation (ZCR-ACR): Used to assess the frequency components of noise. Peak-to-peak ratio (PPR): Used to detect transient impulse interference.

[0026] For example, if a certain data segment has SD>0.5°C and ZCR-ACR>10Hz, it is determined to be high-frequency mechanical vibration noise; if PPR suddenly increases to 3 times the average value, it is determined to be transient thermal shock during material extrusion.

[0027] Window size dynamic adjustment strategy Based on real-time computational NSF, the filter applies a window size adaptation policy (WSAP): When high-frequency noise dominates (ZCR-ACR>threshold TH1, such as 15Hz): Increase the window size (e.g., from 50ms to 100ms) to enhance the low-pass filtering effect and suppress high-frequency noise; When impulse noise bursts (PPR>threshold TH2, such as 2.5): temporarily switch to Median Filter Mode (MFM), with the window size set to 10ms to quickly suppress outliers; In steady-state conditions (SD < threshold TH3, e.g., 0.2°C): reduce the window size to 20ms to preserve true temperature details.

[0028] The adjustment decision is implemented by a fuzzy rule engine (FRE), with input variables being normalized SD, ZCR-ACR, and PPR, and output being the window size increment (WSI).

[0029] Filtering and Denoising Sequence Generation After determining the window size, the filter performs a weighted moving average (WMA): Each sampling point within the window is assigned a Gaussian weight (GW) based on its temporal position relative to the current point, with the center point having the highest weight and decreasing weight at the edges. The data points within the window are weighted, summed, and normalized to output the filtered temperature value (FTV) at the current moment. The window slides in steps of 1 (i.e., every millisecond) to generate a denoised temperature sequence (DTS) point by point.

[0030] This process continues in a loop, recalculating the NSF and updating the window size every 200 milliseconds to ensure that the filtering parameters always match the current noise environment. The output DTS data rate remains at 1kHz, but non-stationary noise is effectively suppressed (signal-to-noise ratio improved by ≥15dB).

[0031] The noise-reduced temperature sequence is processed by Fast Fourier Transform, and the Hanning window function is dynamically selected for spectrum analysis to output a full-band temperature spectrum. Spectrum Analysis Preprocessing The system performs preprocessing on the denoised temperature sequence (DTS) before performing a Fast Fourier Transform (FFT): Data segmentation: Each frame consists of 512 points (512ms), with an inter-frame overlap of 50% (256ms) to ensure that transient events are captured completely; Detrending: Removes the linear trend component (LTC) from each frame of data to avoid spectral leakage. It uses the least squares method to fit a straight line and subtract it. Dynamic window function selection: Automatic window function selection based on the stationarity index (SI) of intra-frame data. If SI > 0.8 (highly stable), select a rectangular window to maximize frequency resolution; If 0.3 ≤ SI ≤ 0.8 (moderate fluctuation), use the Hanning window to balance resolution and leakage suppression; If SI < 0.3 (severe fluctuations), use a flat-top window to ensure amplitude accuracy.

[0032] The Hanning window (with window function coefficients of 0.5 - 0.5×cos(2πn / N)) has become the default choice due to its moderate main lobe width (3dB bandwidth for 1.5 frequency points) and fast side lobe attenuation (-32dB / oct).

[0033] Real-time FFT computation and spectrum generation The preprocessed frame data is input into a fixed-point accelerated FFT processor (such as the FPU unit of an ARM Cortex-M7) to execute the radix-2 FFT algorithm. Calculate a 512-point complex FFT (output 256 valid frequency lines); Frequency domain resolution Δf = sampling rate / number of points = 1000Hz / 512 ≈ 1.95Hz; Calculate the magnitude spectrum (MS): Take the modulus of the complex result and normalize it to a 0 dB reference; Window Function Energy Compensation (WFEC): The Hanning window needs to be multiplied by a correction factor of 2.0; Output Single-Frame Spectrum (SFS), which contains the amplitude (in dB) of 256 frequency points in the range of 0-500Hz.

[0034] Full-band spectrum construction and updating The full-band temperature spectrum (FBTS) is dynamically generated through the following steps: Spectrum stitching: Arrange consecutive frames of SFS in chronological order into a two-dimensional matrix (row = time axis, column = frequency axis, value = amplitude dB). Time-frequency smoothing: A 3-point moving average is applied along the time axis, and a Gaussian kernel (σ=1.5 frequency point) is applied along the frequency axis to suppress random fluctuations; Dynamic range compression: Log compression (LC) is applied to the amplitude to enhance the visibility of weak components (e.g., 70dB dynamic range is compressed to 0-1). Visualization output: The matrix is ​​mapped to a pseudo-color map, where warm colors (red) represent high energy and cool colors (blue) represent low energy, updated 4 times per second (4fps).

[0035] This spectrum diagram visually demonstrates the frequency domain characteristics of temperature fluctuations (such as 50Hz power frequency interference, 100-200Hz fan vibration, and <0.5Hz slow change in thermal inertia), providing a basis for subsequent component separation.

[0036] Based on the full-band temperature spectrum, an energy accumulation algorithm is applied to automatically determine the 0.5Hz segmentation threshold and output the separated low-frequency steady-state component and high-frequency transient component.

[0037] Band energy accumulation and feature extraction The system performs an energy accumulation algorithm (EAA) on the full-band spectrum (FBTS): Band-division integration: Dividing the spectrum into three key frequency bands: Ultra-low band (ULB): 0-0.1Hz, reflecting ambient temperature drift; Main Control Band (MCB): 0.1-10Hz, corresponding to the heater PID control frequency; High Noise Band (HNB): 10-500Hz, including mechanical vibration and electrical noise.

[0038] Energy calculation: Numerical integration (trapezoidal method) is performed on the spectral amplitude of each frequency band within the latest time window (e.g., 2 seconds) to obtain the total band energy (BE), in dB·Hz; Energy Ratio Calculation: Calculating the Critical Energy Ratio (CER): CER1 = BE_ULB / BE_MCB (reflects the intensity of environmental disturbance); CER2 = BE_HNB / BE_MCB (reflects the degree of high-frequency noise pollution).

[0039] Dynamic decision-making for segmentation threshold Based on energy characteristics, the system automatically determines a 0.5Hz segmentation threshold (Dynamic Threshold Decision, DTD): Basic threshold setting: The initial segmentation threshold is set to 0.5Hz (empirical value), dividing the temperature signal into: Low-Frequency Steady-State Component (LFSSC): <0.5Hz, containing slowly varying temperatures dominated by thermal inertia; High-Frequency Transient Component (HFTC): ≥0.5Hz, containing fast-changing disturbances.

[0040] Adaptive adjustment rules: If CER1 > 0.3 (strong environmental disturbance), lower the threshold to 0.2 Hz and expand the LFSSC range to enhance anti-drift capability; If CER2 > 0.4 (strong high-frequency noise), increase the threshold to 1.0 Hz to prevent HNB energy from seeping into LFSSC; If the printhead is moving at high speed (according to G-code speed > 100 mm / s), temporarily increase the threshold to 2.0 Hz to match dynamic thermal load changes.

[0041] The adjusted threshold (AT) is smoothly transitioned through a first-order low-pass filter (time constant 5 seconds) to avoid frequent jumps.

[0042] Component separation and output Finally, frequency domain signal separation is performed: Frequency domain filtering: Applying a zero-phase digital filter to the noise-reduced temperature series (DTS): LFSSC approach: Design an IIR Butterworth low-pass filter (Butterworth LPF) with a cutoff frequency of AT (e.g., 0.5Hz) and an order of 4. HFTC path: Synchronous design of high-pass filters (HPF) with the same cutoff frequency.

[0043] Temporal reconstruction: The DTS is processed through LPF and HPF to generate two time-domain signals respectively; Forward-backward filtering is used to eliminate phase delay and ensure that components are aligned with the original timing sequence. Output interface: LFSSC: Basic temperature tracking for PID controllers; HFTC: Input SNN model to predict transient compensation amount.

[0044] The separation effect is evaluated online using the component orthogonality index (OI) (OI > 20dB is required). If the target is not met, the threshold is readjusted.

[0045] This method acquires real-time temperature information of the printhead through high-frequency sampling and uses frequency domain analysis to decompose the complex temperature fluctuation signal into low-frequency components reflecting long-term trends and high-frequency components reflecting instantaneous changes. This decomposition method can more accurately identify different characteristics of temperature changes, providing a data foundation for subsequent targeted control. By separating the different frequency components of the temperature signal, targeted control strategies can be designed. The low-frequency components reflect the overall thermal equilibrium state of the printhead, while the high-frequency components capture the detailed characteristics of rapid disturbances. This decomposition significantly improves the precision and response speed of temperature control.

[0046] S202, based on the type of printing material and the ambient temperature, predict the thermal response characteristic curve of the material and generate a thermal response compensation coefficient, wherein the prediction is based on a mapping relationship library of material specific heat capacity, thermal conductivity and ambient temperature; Specifically, the material type code for printing can be obtained through a material identification sensor, while the ambient temperature sensor data can be read to output the material type identifier and the real-time ambient temperature value. When the system initiates the temperature control optimization process, it first needs to accurately identify the type of printing material currently in use and monitor the ambient temperature. Material type identification is accomplished through a dedicated Material Identification Sensor (MIS). This sensor is typically integrated near the printer's material loading mechanism (such as the feed throat or extruder inlet), and its core technology may employ Near Field Communication (NFC) tag reading, optical barcode / QR code scanning, or Radio Frequency Identification (RFID) technology based on the material's dielectric properties. For example, when a user loads a roll of PLA (polylactic acid) plastic filament into the printer, the embedded NFC tag (NFCT) on the material roll is activated by the antenna of the Material Identification Sensor (MIS). The NFCT stores a globally unique Material Type Code (MTC), such as "PLA-STD-WHITE-1.75". After reading this code, the MIS decodes it into a system-recognizable digital signal, generating a Material Type Identifier (MTI). Meanwhile, a high-precision ambient temperature sensor (ATS) installed inside the printer housing or near the control board operates continuously. The ATS typically uses a digital temperature chip (such as the DS18B20 or TMP117) to read the real-time ambient temperature value (RATV) of the surrounding air at a sampling rate (SR) of 1-10 times per second via a single-wire or I²C interface, usually in degrees Celsius (°C). For example, in a closed printing chamber, the ATS might read a temperature of 32.5°C. These two key data points—MTI (e.g., "PLA-STD-WHITE-1.75") and RATV (e.g., 32.5)—are packaged into a data packet and passed as the output of this step to subsequent processing modules. This process ensures that the system accurately knows the two key prerequisites: "what material is being used" and "how hot the current environment is."

[0047] The reliability of the Material Identification Sensor (MIS) is crucial. To address potential identification failures (such as damaged labels, contamination, or unlabeled new materials), the system is designed with multiple backup mechanisms. When the MIS cannot read a valid Material Type (MTC), it triggers the Auxiliary Identification Procedure (AIP). The AIP first attempts to generate an MTI based on the user's manual selection of the material type on the printer control interface (e.g., a touchscreen). If the user does not interact, the AIP may activate the extruder for a small feed movement (e.g., 0.5 mm), while monitoring changes in the extrusion motor torque (TQ) or current (CUR). Different materials (e.g., soft TPU versus rigid ABS) require different TQ / CUR values ​​at the same feed rate. The system's built-in Material Mechanical Property Library (MMPL) stores typical TQ / CUR ranges for common materials. By matching the measured TQ / CUR with the characteristic values ​​in MMPL, the system can infer the possible material type, generate a temporary MTI (such as "suspected - TPU"), and prompt the user for confirmation. Data from the ambient temperature sensor (ATS) needs to consider interference from local heat sources (such as printer motherboard heat dissipation). The ATS is typically installed away from heat sources in well-ventilated locations and may be equipped with a mini fan (MF) for forced convection. Furthermore, ATS readings are processed using a moving average filter (MAF), such as calculating the average value (AVG) of the most recent 10 readings, to eliminate transient fluctuations (such as brief temperature changes caused by door opening), resulting in a more stable RATV output. All identification and acquisition events (including success, failure, and backup triggering) are logged in the system log (SYSLOG) for maintenance reference. Through these measures, the system ensures the accuracy, stability, and robustness of MTI and RATV data, laying a solid foundation for subsequent thermal property prediction.

[0048] For scenarios supporting multi-material printing (such as dual extruder models) or frequent material changes, the system needs to possess Dynamic Tracking Capability (DTC). When a material change action is detected (such as an extruder unloading / loading signal or user interface operation), the system immediately re-triggers the data acquisition process of the Material Identification Sensor (MIS) and Ambient Temperature Sensor (ATS) to ensure that the MTI and RATV always reflect the current actual state. Simultaneously, the system maintains an Active Material Context (AMC), which not only contains the current MTI and RATV but also records relevant timestamps (TS), sensor confidence levels (SCL), and any auxiliary identification information. The AMC, as a global data structure, is shared and accessed by all subsequent modules that depend on material type and ambient temperature. For example, when the printer switches from printing PLA to printing PETG, the MIS reads the new MTI (e.g., "PETG-TRANSP-BLUE-1.75"), the ATS updates the RATV (which may become 30.1°C due to the door opening caused by the material change), and the AMC updates accordingly. This dynamic update mechanism ensures that the entire temperature control system can respond to changes in materials and the environment in real time, avoiding control inaccuracies caused by using outdated information. MTI and RATV data also undergo format validation (FV) before output to ensure they conform to predefined data structures (e.g., MTI as a string, RATV as a floating-point number), preventing invalid data from flowing into downstream modules.

[0049] Input the material type identifier into the material thermal property database, retrieve the corresponding specific heat capacity and thermal conductivity benchmark values, and output the material basic thermal property vector. After obtaining the Material Type Identifier (MTI), the system needs to query its inherent thermophysical properties (TPP), the core of which are specific heat capacity (SHC) and thermal conductivity (THC). These properties are stored in the Material Thermal Property Database (MTPDB) in the printer firmware or a cloud server. The MTPDB is a structured database (such as SQLite or an embedded key-value store), and its core entries use the MTI as the primary key (PK). Each entry contains at least two key fields: SHC_B (reference specific heat capacity, unit: joules per kilogram Kelvin, J / (kg·K)) and THC_B (reference thermal conductivity, unit: watts per meter Kelvin, W / (m·K)). For example, for MTI="PLA-STD-WHITE-1.75", the database might store SHC_B=1800 J / (kg·K) and THC_B=0.13 W / (m·K); for MTI="ABS-HI-IMPACT-BLACK-1.75", it would store SHC_B=1600 J / (kg·K) and THC_B=0.18 W / (m·K). The system uses the received MTI (such as "PLA-STD-WHITE-1.75") as the query key (QK) and sends a query request (QR) to the MTPDB. The database engine executes an exact match query (EMQ) to retrieve the SHC_B and THC_B values ​​corresponding to that MTI. The query result (QRST) is encapsulated into a data structure containing two floating-point values, called the Material Basic Thermal Property Vector (MBTPV), typically in the form [SHC_B, THC_B]. For example, the query result would be [1800.0, 0.13].

[0050] The database query process needs to consider various anomalies. If the input MTI cannot be precisely matched in MTPDB (e.g., the user used an unregistered new material or the MTI was entered incorrectly), the system will initiate the Similar Material Matching Strategy (SMMS). SMMS first attempts to perform fuzzy matching (FM) based on the material name, for example, ignoring color and diameter information (simplifying "PLA-STD-WHITE-1.75" to "PLA-STD"), to find the closest generic type entry in the database (e.g., "PLA-GENERIC"). If no result is found, SMMS searches the database for the default value (DEF) of the same category of material based on the material mechanical properties (e.g., torque TQ) obtained from the previous Assisted Identification Process (AIP) or the material category manually selected by the user (e.g., "PLA"). As a last resort, the system can connect to the cloud database (CDB) for online querying or prompt the user to manually enter the SHC_B and THC_B values. All values ​​obtained through non-exact matching are labeled as estimated values ​​(EST) and their subsequent confidence weight (CW) is reduced. The database itself is updated periodically (DBUP) by downloading new material thermal property data packages (DP) released by manufacturers via printer network or by allowing users to input SHC_B and THC_B values ​​for custom materials through the Calibration Tool (CALTOOL). Before being output as MBTPV, the retrieved SHC_B and THC_B values ​​undergo a reasonableness check (RC), such as checking whether SHC_B is within the typical range (1000-2500 J / (kg·K)) for common plastics, to prevent erroneous data from causing subsequent prediction failures.

[0051] MBTPV (e.g., [1800.0, 0.13]) includes not only the numerical value itself but also metadata (MD). The metadata includes the data source (exact match, fuzzy match, estimate), confidence level (CL, e.g., high / medium / low), timestamp (TS), and the associated original MTI. This metadata is crucial for subsequent modules, especially the thermal response prediction model. For example, if the MBTPV originates from a high-confidence exact match, the thermal response prediction model can use these baseline values ​​more aggressively; if it's a low-confidence estimate, the model may need to rely more heavily on real-time calibration or adopt a more conservative control strategy to avoid risk. The system also maintains a Thermal Property Cache (TPC), storing recently used MTIs and their corresponding MBTPVs (along with metadata) in memory. When the same MTI is repeatedly queried within a short period (such as in continuous print jobs), the system prioritizes reading MBTPV from the TPC cache, significantly reducing database access latency (LR) and improving system real-time performance. The MBTPV output interface is designed in a standardized format to ensure strict compatibility with the input requirements of the next processing step.

[0052] Input the material's basic thermal property vector and the ambient temperature value into the thermal response prediction model, calculate the temperature compensation offset, and output the preliminary thermal response curve. After obtaining the material's fundamental thermal property vector (MBTPV = [SHC_B, THC_B]) and real-time ambient temperature value (RATV), the system needs to predict the material's thermal response characteristics at the current ambient temperature. This is achieved through the Thermal Response Prediction Model (TRPM). TRPM is a mathematical model built based on physical principles and experimental data, with its core inputs being MBPPV (providing the material's inherent properties) and RATV (providing external environmental conditions). The model first recognizes that the material's actual thermal behavior (such as heating and cooling rates) depends not only on its baseline specific heat capacity (SHC_B) and baseline thermal conductivity (THC_B), but is also significantly affected by the ambient temperature (RATV). For example, for the same PLA material, the required heating power and response time to reach the same printhead target temperature (e.g., 210°C) will differ between a cold environment of 15°C and a hot environment of 35°C. The core task of the model is to calculate a Temperature Compensation Offset (TCO) based on the inputs, which will be used to correct the behavior predicted based on the baseline properties. TRPM may contain a set of empirical formulas or lookup tables (LUTs). A simplified calculation example: TCO = K1 * (RATV - Tref) * SHC_B + K2 * (RATV - Tref) * (1 / THC_B). Where K1 and K2 are model coefficients (MCs), calibrated through extensive thermal testing of materials under various environments; Tref is the reference ambient temperature (e.g., 25°C). This formula reflects the degree to which the material needs to compensate for changes in specific heat capacity (affecting heat absorption capacity) and thermal conductivity (affecting heat transfer rate) when the ambient temperature deviates from the reference value.

[0053] Using the calculated Temperature Compensation Offset (TCO), the model further generates a Preliminary Thermal Response Curve (PTRC). This curve describes the expected dynamic behavior of the Printer Head Temperature (PHT) over time or with varying input power at a given ambient temperature (RATV). The curve is typically presented in the form of a First-Order System Model (FOSM): PHT(t) = PHT_ss + (PHT_initial - PHT_ss) * exp(-t / τ) + TCO_effect. Here, PHT_ss is the steady-state target temperature (determined by print settings), PHT_initial is the initial temperature (close to RATV), t is time, and τ is the time constant (TC), characterizing the system response speed. The calculation of τ is usually related to MBTPV, for example, τ = C * Mass / (THC_B * Area) (where C is a constant, Mass is the printhead thermal mass, and Area is the effective heat transfer area). TCO_effect represents the impact of TCO on the curve shape, which may manifest as an offset of PHT_ss (static offset) or a correction of τ (dynamic response speed change). The model generates the complete PTRC by analytical calculation or interpolation of pre-stored curve templates. This curve is output as discrete data points (DP), such as a sequence (SEQ) of temperature predictions uniformly sampled (e.g., one point per second) along the time axis, or as characteristic parameters (e.g., τ, steady-state gain). PTRC serves as the basis for subsequent control and calibration predictions.

[0054] The complexity and accuracy of TRPM can be adjusted based on system resources. In resource-constrained embedded systems, TRPM may be implemented as a simplified computation based on polynomial regression (PR). In more powerful systems, a more refined finite difference thermal model (FDTM) may be used, discretizing the printhead structure into a grid and combining SHC_B and THC_B from MBTPV with RATV as boundary conditions (BC) to simulate the heat conduction process, outputting a more accurate but computationally more computationally intensive PTRC. The model also considers the material's thermal history (TH). If the system detects that the material has just undergone high-intensity printing (printhead at high temperature) or has been idle for a long time (printhead cooling to ambient temperature), TRPM may fine-tune its predictions. For example, for material that has just cooled down from a high temperature, its initial temperature PHT_initial may be higher than RATV, and the model will adjust the starting point of PTRC accordingly. PTRC performs a smoothing process (SP) before output, such as applying moving averages or low-pass filtering, to eliminate numerical fluctuations that may occur during calculation, ensuring a smooth and continuous curve for easier processing by subsequent modules. Parameters such as model coefficients (MC) and reference temperature (Tref) are stored in configurable non-volatile memory (NVM), allowing for adjustment and optimization through firmware updates or user calibration.

[0055] The initial thermal response curve is input into the real-time calibration module, and the curve parameters are corrected by combining historical temperature fluctuation data to output an optimized thermal response curve. The preliminary thermal response curve (PTRC) is a prediction based on theoretical models and baseline parameters, which may deviate from the actual thermal response during actual printing. To narrow this gap, the system introduces a real-time calibration module (RTCM). The core function of the RTCM is to correct the PTRC parameters using historical temperature fluctuation data (HTFD) collected by the system during recent actual printing processes. HTFD is derived from continuous records from high-precision temperature sensors (such as thermocouples or thermistors) installed on the printhead. The system maintains a circular buffer (CB) that stores a sampling sequence of the actual head temperature (AHT) over a recent period (e.g., the past 5-10 minutes) as a function of time, while also recording the corresponding heating power command (HPC). This data forms a valuable record of the actual system response.

[0056] After receiving the PTRC, the RTCM performs a dynamic comparison (DC) with the stored HTFD. The comparison is typically performed within the same target setpoint (TSP) time period. The core of the calibration algorithm (CA) is to calculate the error (ERR) between the temperature trajectory predicted by the PTRC and the actual trajectory recorded by the HTFD. Key errors of concern include: steady-state error (SSE, the difference between the predicted and actual steady-state temperatures), rise time error (RTE, the difference between the predicted time to reach the set temperature and the actual time), and overshoot error (OE, the difference between the maximum predicted temperature exceeding the set value and the actual overshoot). For example, the algorithm might find that for the current material at 35°C, the PTRC predicts a rise time of 8 seconds, but the HTFD shows an actual average rise time of 10 seconds (a negative RTE), and the actual overshoot is 3°C higher than predicted (a positive OE). Based on these error analyses, CA employs optimization algorithms (such as the Least Squares Method (LSM) or Gradient Descent (GD)) to back-calculate and correct the parameters of the PTRC. The most frequently corrected parameters are the time constant τ and the steady-state gain G (or the equivalent power-temperature conversion factor). For example, if the actual response is consistently slower than the prediction (negative RTE), CA increases the value of τ in the PTRC; if the steady-state temperature is consistently lower than the prediction (negative SSE), CA increases the value of G. The magnitude of the correction is proportional to the observed error magnitude and confidence level (based on the amount of HTFD data). The correction process is iterative and gradual, avoiding drastic parameter jumps caused by a single anomalous data point.

[0057] The curve after parameter correction becomes the Optimized Thermal Response Curve (OTRC). OTRC not only more accurately reflects the true thermal behavior of the material under current conditions but also implicitly includes the impact of the current printhead physical state (such as a slight decrease in heating rod efficiency or changes in thermal resistance at the measurement point). The RTCM output (OTRC) is also represented as a discrete point sequence or characteristic parameters. The system is designed with a Calibration Confidence Assessment (CCA). If the HTFD data is sufficient (e.g., covering the complete heating, steady-state, and cooling processes) and the data quality is high (low noise), the corrected OTRC confidence level (CL) is marked as high; conversely, if the data is insufficient or of poor quality, the OTRC confidence level is low, and the system may rely more on the original PTRC or prompt the user to perform manual calibration. The calibration parameters and the OTRC itself also have time sensitivity. The system periodically resets or decays previous calibration results, restarting the learning process and ensuring model adaptability, either when a significant change is detected (such as printhead replacement or restart after prolonged inactivity) or when it is detected. The OTRC is ultimately passed to the next module for extracting key control parameters.

[0058] The time constant and gain factor are extracted from the optimized thermal response curve, and the normalization coefficient is calculated by weighted fusion to output the thermal response compensation coefficient.

[0059] The optimized thermal response curve (OTRC) contains rich dynamic information, but subsequent compensation control algorithms (such as the spiking neural network (SNN) as described in claim 4) typically require simpler and more representative feature parameters. Therefore, the core of this step is to extract two key features from the OTRC: the time constant τ and the gain factor G. The time constant τ (unit: seconds) quantitatively describes the system's (printhead + material) response speed to changes in heating power. It can be obtained by analyzing the rise phase (time required to reach 63.2% of the target temperature from the initial temperature) or the fall phase (time required to decrease from the steady-state temperature to 36.8%) of the OTRC. The gain factor G (unit: °C / Watt) describes the system's steady-state characteristics, representing the steady-state temperature rise achievable per unit heating power input. It can be obtained by calculating (target steady-state temperature - ambient temperature RATV) / the average heating power required to reach that steady state. For example, OTRC analysis yields τ = 5.2 seconds and G = 0.8 °C / W. These two parameters (τ, G) together define the core dynamic behavior of OTRC.

[0060] The extracted τ and G need to be fused into a single, dimensionless Thermal Response Compensation Coefficient (TRCC) for use by subsequent modules. The fusion process employs weighted fusion (WF). The system assigns a weight factor (WF) to τ and G respectively: W_τ and W_G. The weight allocation is determined based on empirical rules or optimization, typically reflecting the relative importance of the two parameters to the final compensation effect. For example, in scenarios with drastic temperature fluctuations, dynamic response speed (τ) may be more important, so W_τ is set larger (e.g., 0.7); in scenarios where steady-state accuracy is prioritized, gain (G) may be more important, so W_G is set larger (e.g., 0.6). Fusion calculations usually involve a normalization (NRM) step because the dimensions and numerical ranges of τ and G differ significantly. A common fusion formula is: TRCC_raw = W_τ * (τ_ref / τ) + W_G * (G / G_ref). Here, τ_ref and G_ref are reference values ​​(REF), typically chosen as the typical values ​​of a standard material (such as PLA) under standard conditions (such as 25℃) (e.g., τ_ref = 4.0 seconds, G_ref = 1.0 ℃ / W). This treatment makes TRCC_raw dimensionless, and when τ = τ_ref and G = G_ref, TRCC_raw ≈ 1. In the formula, the τ_ref / τ term means that the faster the response (the smaller τ), the greater the contribution of this term; the G / G_ref term means that the greater the gain (the larger G), the greater the contribution of this term.

[0061] The calculated TRCC_raw (e.g., 1.15 or 0.92) typically requires normalization (NP) and range limiting (RL) to output the final TRCC. Normalization ensures that TRCC falls within a standard range (e.g., 0.5 to 2.0) for easier processing by subsequent algorithms. One approach is to determine the typical range of TRCC_raw based on a large number of samples and then perform linear scaling (LS): TRCC = a * TRCC_raw + b, where a and b are scaling factors. Another approach is to apply a nonlinear transformation using the Sigmoid function (SF) to map TRCC_raw to the (0,1) or (0.5, 1.5) interval. Range limiting prevents extreme values ​​(e.g., sensor malfunctions causing errors in τ or G calculations) from causing subsequent control failures; for example, it forces TRCC to be between [0.5, 2.0]. The final output TRCC is a floating-point number (FPN), such as 1.08. This coefficient encapsulates the combined effects of material properties and ambient temperature on the printhead's thermal response: TRCC > 1 typically indicates that the system is more difficult to heat up or responds more slowly than the reference state (requiring stronger compensation), while TRCC < 1 indicates that it is easier to heat up or responds more quickly (requiring weaker compensation). TRCC is output along with its source OTRC confidence information (CL), serving as a key input for subsequent dynamic compensation power calculations. The system records the calculation process, input parameters, and final value of TRCC for each iteration, used for performance analysis and continuous model optimization.

[0062] This method comprehensively considers material properties and environmental conditions, establishing a predictive model of material thermal behavior under specific environments through a pre-set database of thermophysical parameters. The system can dynamically generate compensation parameters reflecting the material's thermal properties based on the currently used printing material and the measured ambient temperature. This prediction method based on physical properties effectively solves the control deviation problem caused by differences in the thermal response of different materials, enabling the temperature control system to adapt to various printing materials and significantly improving the consistency of temperature control during multi-material printing.

[0063] S203, a PID controller is used to generate the basic heating power for the low-frequency steady-state component, and the amplitude feature vector is extracted from the high-frequency transient component. The amplitude feature vector and the thermal response compensation coefficient are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. Specifically, the low-frequency steady-state component can be input into an adaptive PID controller, which will automatically adjust the control parameters according to the temperature change rate and output a basic heating power signal. The system receives the low-frequency steady-state component (LF-SSC) separated by a fast Fourier transform. This component mainly reflects the slow temperature change trend of the printhead (such as ambient temperature drift and long-term heat accumulation). The core function of the adaptive PID controller (A-PID) is to dynamically adjust the three parameters: proportional (P), integral (I), and derivative (D). The controller first calculates the real-time temperature change rate (TCR) of the LF-SSC, in degrees Celsius per second (°C / s). For example, if the temperature rises from 200.1°C to 200.15°C within two consecutive sampling periods (sampling rate 1kHz, period 1ms), the TCR is 50°C / s. Based on the absolute value and direction (heating / cooling) of the TCR, the controller calls the preset parameter tuning rule base (PTRB). For example, when the TCR is detected to be greater than 100°C / s (rapid heating), the proportional gain P is automatically reduced (e.g., from 5.0 to 3.0) to avoid overshoot; when the TCR approaches 0°C / s (steady-state maintenance), the integral gain I is increased (e.g., from 0.05 to 0.1) to eliminate steady-state error. This real-time parameter adjustment ensures that the PID controller can adapt to the dynamic characteristics of different operating conditions.

[0064] The PID controller output employs discretized iteration: the controller takes the deviation (Error, E) between the current temperature setpoint (TargetTemperature, TT, e.g., 210°C) and the actual LF-SSC value, the historical cumulative deviation (Integral of Error, IE), and the current rate of change of deviation (Derivative of Error, DE) as inputs. For example, if the current E is -5°C (actual temperature is lower than the target), IE is -100°C / s (continuously low), and DE is -0.2°C / s (deviation is still increasing), the controller calculates the output using the adjusted P / I / D parameters. After each calculation, the values ​​of IE and DE are updated in the register for use in the next cycle. To prevent integral saturation, the controller introduces a conditional integration (CI) mechanism: IE is accumulated only when the error E is within a preset threshold (e.g., ±10°C); otherwise, the integral term is frozen. The final output Base Heating Power Signal (BHPS) is an analog quantity ranging from 0 to 100 percent (%), representing the percentage of full power supplied by the heating element. For example, BHPS=75% means that 75% of the maximum rated power needs to be applied.

[0065] The controller's adaptive capabilities are further enhanced by the Historical Operation Learning Module (HOLM). This module continuously records the optimal P / I / D parameter combinations and control effects (such as settling time and overshoot) under different TCR ranges. When a similar operating condition is detected to recur (e.g., TCR stabilizes at 30±5°C / s), HOLM automatically fine-tunes the parameter baseline values ​​in PTRB. For example, if historical data shows that increasing the differential gain D to 2.5 under this condition reduces temperature fluctuation by 5%, the new parameter will override the old value. Simultaneously, the controller has a built-in Abnormal Condition Handler (ACH). When a sudden TCR change exceeds the safety threshold (e.g., >500°C / s), it immediately switches to a conservative parameter preset group (e.g., P=1.0, I=0.01, D=0.5) and triggers an alarm. The entire adaptive PID process runs in real-time at a frequency of 1kHz, ensuring that the BHPS accurately matches the slow heating requirements of the printhead.

[0066] An envelope detection algorithm is applied to the high-frequency transient components to extract peak value, root mean square and zero-crossing rate features, and output amplitude feature vector. For the High-Frequency Transient Component (HF-TC), the system employs an Envelope Detection Algorithm (EDA) to capture its oscillatory characteristics. This algorithm first constructs the analytic signal (AS) of the HF-TC using the Hilbert Transform (HT), from which the instantaneous amplitude envelope (IAE) is extracted. For example, an oscillating signal with an amplitude of ±2°C and a frequency of 50 Hz has an IAE that is a smooth curve with an amplitude of 2°C. After the IAE is generated, the system calculates three key features using a sliding window of 10 milliseconds (ms): Peak Value (PV): The maximum value of the IAE within the window, reflecting the intensity of extreme temperature fluctuations. For example, PV = 1.8°C indicates that the highest transient offset within the window is 1.8°C.

[0067] Root Mean Square (RMS): The root mean square value of the IAE within the window, representing the average fluctuation energy. For example, RMS = 1.2°C means the effective value of the fluctuation is 1.2°C.

[0068] Zero-Crossing Rate (ZCR): The number of times the IAE crosses its mean within a window, indicating the oscillation frequency. For example, ZCR = 25 times / 10ms is equivalent to an effective frequency of 250 Hz.

[0069] The feature extraction process needs to address high-frequency noise interference. Before envelope detection, the system performs wavelet threshold denoising (WTD) on the HF-TC signal: a 3-level decomposition using the Daubechies4 wavelet basis is performed, and soft thresholding is applied to detail coefficients to remove noise components with amplitudes less than 0.1°C. The denoised signal is then input into the EDA to ensure the accuracy of PV, RMS, and ZCR. After calculation, the three feature values ​​are combined in a fixed order (PV→RMS→ZCR) to form an amplitude feature vector (AFV). For example, a window outputs AFV=[1.8, 1.2, 25], with a three-dimensional (3D) vector. The AFV sampling rate is consistent with the window sliding step size (100 Hz), meaning 100 feature vectors are generated per second.

[0070] To adapt to different fluctuation patterns, the system introduces Dynamic Window Adjustment (DWA). When the ZCR is detected to be consistently higher than 500 Hz (e.g., ultra-high frequency noise), the window size is automatically shortened from 10 ms to 5 ms to improve temporal resolution; when the ZCR is lower than 10 Hz (e.g., low-frequency interference), the window is expanded to 20 ms to smooth out occasional fluctuations. Simultaneously, before outputting the AFV, feature normalization (FN) is performed: based on the maximum PV (e.g., 3.0°C), maximum RMS (e.g., 2.0°C), and maximum ZCR (e.g., 500 Hz) statistically derived from historical data, each dimension of the vector is scaled to the [0,1] interval. For example, AFV=[1.8,1.2,25] is normalized to [0.6, 0.6, 0.05]. The normalized AFV serves as the standardized input for subsequent neural networks.

[0071] The amplitude eigenvector and the thermal response compensation coefficient are concatenated to generate an enhanced eigenvector. The fusion of the Amplitude Eigenvector (AFV) and the Thermal Response Compensation Coefficient (TRCC) employs a Feature Concatenation Operation (FCO). TRCC is a scalar value (SV) generated by the preceding module, typically ranging from 0.5 to 2.0 (dimensionless), representing a sensitivity correction factor for the current printed material to changes in ambient temperature. For example, TRCC=1.2 indicates that the material's thermal response is 20% faster than the standard material. The concatenation operation directly concatenates the three-dimensional vector of AFV (PV_norm, RMS_norm, ZCR_norm) with TRCC to form a four-dimensional (4D) Enhanced Feature Vector (EFV). For example, if AFV=[0.6, 0.6, 0.05] and TRCC=1.2, then EFV=[0.6, 0.6, 0.05, 1.2].

[0072] Before cascading, the difference between the units and numerical scales needs to be addressed. Since the AFV has been normalized to [0,1], while the physical range of TRCC is [0.5,2.0], the system performs linear scaling (LS) on TRCC: mapping it to the [0,1] interval. The mapping formula is: TRCC_norm = (TRCC - 0.5) / (2.0 - 0.5). For example, TRCC=1.2 is scaled to TRCC_norm≈0.47 ((1.2-0.5) / 1.5). The scaled TRCC_norm is then concatenated with the AFV to ensure that the numerical ranges of each dimension of the EFV are consistent. In addition, the system sets an outlier filter (OF): if TRCC is detected to be outside the preset range (e.g., <0.3 or >2.5), the current EFV generation is frozen and the calibration process is triggered to prevent erroneous data from contaminating subsequent models.

[0073] To enhance feature representation capabilities, EFV injects contextual features (CF) after generation. These include: Printhead Status Flag (HSF): Binary value (0 / 1), 0 indicates idle state, 1 indicates heating state; Historical Fluctuation Trend (HFT): The RMS mean of the previous 10 AFVs, reflecting the intensity of recent volatility.

[0074] For example, when HSF=1 and HFT=0.7, the EFV expands from [0.6,0.6,0.05,0.47] to a six-dimensional (6D) vector [0.6,0.6,0.05,0.47,1,0.7]. The added dimension is dynamically enabled via a Feature Selector (FS). HSF and HFT are automatically loaded when the printhead is in high-power mode (e.g., BHPS>80%); otherwise, only the basic four-dimensional EFV is retained. Finally, the EFV is output to the downstream neural network at a frequency of 100 Hz.

[0075] The enhanced feature vector is input into a pre-trained spiking neural network model, and the original compensation power is output through spatiotemporal event-driven processing. The pre-trained Spiking Neural Network (SNN) model employs a three-layer topology: an input layer (6 neurons corresponding to the EFV dimension), a hidden layer (32 Leaky Integrate-and-Fire (LIF) neurons), and an output layer (1 linear decoding neuron). Before inputting into the EFV, it needs to be converted into a spike train (ST): each feature value is converted into a spike frequency through rate coding (RC). For example, a normalized value of 0.6 in the EFV corresponds to a 60 Hz spike stream (i.e., one spike every 16.7 ms). After receiving the spikes, the input layer neurons update the membrane potential (MP) according to the LIF model: when MP exceeds a threshold (e.g., 30 mV), a spike is triggered and MP is reset. The spike outputs of the hidden layer neurons are transmitted to the output layer through the Synaptic Weight Matrix (SWM).

[0076] The core of Spatio-Temporal Event-Driven Processing (STEDP) is the dynamic adjustment of synaptic weights. The system monitors the fluctuation frequency represented by ZCR_norm in the EFV in real time. When ZCR_norm > 0.3 (high frequency fluctuation), activate the High-Frequency Subnet (HFS) and enhance the neuron connection weights that are sensitive to PV and ZCR (e.g., increase the weights by 20%). When ZCR_norm < 0.1 (low-frequency fluctuation), the Energy Accumulation Subnet (EAS) is enabled to increase the weight dependence on RMS and HFT (e.g., the weight is increased by 30%).

[0077] The weight adjustment is based on the Spike-Timing-Dependent Plasticity (STDP) rule learned during pre-training: if the pulse of the preceding neuron precedes that of the following neuron, the weight is increased; otherwise, it is decreased. For example, if the pulse of the PV neuron in the input layer precedes that of the hidden layer neuron by 5 milliseconds (ms), its synaptic weight is increased by 0.01.

[0078] The linear decoding neurons in the output layer sum the weighted values ​​of all input pulses to generate the raw compensation power (RCP). RCP is an analog quantity with positive and negative directions (range -50% to +50%). A positive value indicates that additional power is needed to suppress cooling fluctuations, while a negative value indicates that power needs to be reduced to suppress heating fluctuations. For example, when EFV indicates severe high-frequency cooling fluctuations (PV_norm=0.9, ZCR_norm=0.4) and TRCC_norm=0.6 (material sensitive), the SNN might output RCP=+35%. The SNN's inference latency is controlled within 1 millisecond (ms) to meet real-time requirements. Model weights are generated through offline supervised training (OST): using thousands of samples covering fluctuations in different materials and ambient temperatures, the optimization objective is to minimize the mean squared error (MSE) between the actual and target temperatures.

[0079] The original compensation power is input into the exponential smoothing filter, and the smoothing factor is adjusted according to the frequency of temperature fluctuations to output dynamic compensation power.

[0080] The Exponential Smoothing Filter (ESF) performs noise reduction on the original compensated power (RCP). Its core formula is the smoothing equation: the current output value (S_t) is a weighted average of the previous output value (S_{t-1}) and the current input value (RCP_t), with the weights determined by the smoothing factor (SF). For example, if SF = 0.3, then S_t = 0.7 × S_{t-1} + 0.3 × RCP_t. The initial value of SF is set to 0.5 (for balance, smoothing, and response speed). The filter dynamically adjusts SF based on the temperature fluctuation frequency (TFF): TFF is obtained by converting ZCR_norm in the EFV (e.g., ZCR_norm = 0.05 corresponds to TFF = 25 Hz). The adjustment rule is as follows: TFF < 20 Hz: SF = 0.2 (strong smoothing, suppressing low-frequency noise); 20 Hz ≤ TFF ≤ 100 Hz: SF=0.5 (moderate smoothing); TFF>100 Hz: SF=0.8 (weak smoothing, preserving high-frequency details).

[0081] To prevent response lag during rapid operating condition changes, the filter incorporates a predictive compensation mechanism (PCM). When the rate of change of RCP exceeds 10% / ms for three consecutive cycles (e.g., a sudden increase from -20% to +30%), the SF is temporarily reduced to 0.1 (with almost no smoothing), ensuring the filter output closely follows the RCP abrupt change. Simultaneously, the ESF has built-in boundary protection (BP): when RCP exceeds the range of [-50%, +50%], it automatically clamps to the boundary value to avoid outlier interference. For example, if RCP = +60%, the ESF is forced to be input at +50%.

[0082] The final output Dynamic Compensation Power (DCP) is a stable signal after adaptive smoothing. For example, when the input RCP oscillates violently within ±30% (high-frequency fluctuation), the ESF outputs DCP = ±25% with SF = 0.8, smoothing out some noise spikes while still retaining the main fluctuation trend; when the RCP stabilizes at -10% (low-frequency drift), the ESF outputs a smoothed DCP = -9.8% with SF = 0.2. The DCP and the Base Heating Power Signal (BHPS) are synchronously output to the downstream fusion module at 100 Hz, together forming the final heating command. All parameter adjustments of the ESF are recorded in the log for maintenance personnel to analyze filter performance.

[0083] This method employs a hierarchical control strategy. For low-frequency components, traditional PID control is used to maintain basic temperature stability. Simultaneously, for high-frequency components, feature extraction is performed and input into a neural network to generate compensation quantities for transient changes. The spiking neural network effectively handles timing characteristics, achieving more precise dynamic compensation. This composite control strategy ensures the stability of the basic control while achieving intelligent compensation for rapid disturbances through the neural network, thus solving the problem of lag in response to sudden temperature fluctuations in traditional control methods.

[0084] S204, based on the basic heating power and the dynamic compensation power, generates a final heating command for the printhead, and constrains the command output range through an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature.

[0085] Specifically, the base heating power and dynamic compensation power can be input into the dynamic weighted fusion unit, the weight ratio can be adjusted according to the temperature deviation, and the initial heating command can be output. The Dynamic Weighted Fuser (DWF) is the core module of power synthesis, and its physical implementation typically employs programmable analog circuits or digital signal processors (DSPs). Base Heating Power (BHP), generated by a PID controller, represents the steady-state power required to maintain the target temperature. Dynamic Compensation Power (DCP), derived from a spiking neural network (SNN), is used to counteract high-frequency temperature disturbances. After both are input into the DWF, the system first reads the real-time temperature deviation (TD), which is the difference between the target temperature and the actual temperature (in degrees Celsius, °C). For example, when the absolute value of TD is greater than 5 °C, it indicates that temperature control is in a transient process. In this case, the DWF increases the weight ratio (WR) of DCP, up to a maximum of 70%, allowing the compensation power to dominate the response. When TD is less than 1 °C, the system enters a steady state, and the weight of BHP increases to over 90%, ensuring control stability. Weight adjustment is achieved through the Fuzzy Logic Decider (FLD): This module quantifies TD into five levels: "negative large", "negative medium", "zero", "positive medium" and "positive large", and presets the corresponding WR rule base (e.g., the DCP weight is set to 65% under the "negative large" level).

[0086] Dynamic weight calculations must consider power change trends. The system additionally monitors the first derivatives of BHP and DCP (d_BHP / dt, d_DCP / dt), i.e., the power change rate (unit: watts / second W / s). If d_DCP / dt exceeds the threshold (e.g., 50 W / s) for three consecutive sampling periods, it indicates a sudden thermal disturbance (e.g., printhead contact with the cooling substrate). In this case, even if TD is small, DWF will temporarily increase the DCP weight to 50% to achieve advance compensation. To avoid frequent weight oscillations, DWF has a built-in hysteresis comparator (HC) with a minimum weight holding time (MWHT) of 0.2 seconds. For example, when WR is reduced from 60% to 40%, it must remain stable at 40% for at least 0.2 seconds before further adjustment is allowed. The final weight output must undergo normalization processing (NP) to ensure that the sum of the BHP and DCP weights is always 100%.

[0087] The synthesis of the Preliminary Heating Command (PHC) adopts a linear superposition model: PHC = WR_BHP × BHP + WR_DCP × DCP.

[0088] Where WR_BHP represents the weight ratio of BHP, and WR_DCP represents the weight ratio of DCP. For example, when WR_BHP=0.7 and WR_DCP=0.3, if BHP=100W and DCP=30W, then PHC=0.7×100 + 0.3×30 = 79W. The PHC generation frequency is consistent with the temperature sampling rate (1kHz), and it is output as an analog voltage signal (range 0-5V, corresponding to 0-150W power) through a digital-to-analog converter (DAC) for subsequent module processing.

[0089] The initial heating command is input into the anti-saturation limiter, and the output range is limited by the dual threshold constraint algorithm to output a safe heating command. The core function of the anti-windup limiter (AWL) is to prevent actuator failure caused by instruction out-of-bounds errors. Its hardware consists of a high-speed comparator (CMP) and a clamping circuit (CC). A dual-threshold constraint algorithm sets two critical boundaries: Static Safety Threshold (SST): Determined by the physical limits of the printhead, for example, an upper limit of 120W (corresponding to a maximum temperature resistance of 300℃ for the heating element) and a lower limit of 0W.

[0090] Dynamic Regulation Threshold (DRT): Adjusted dynamically based on the historical rate of change of commands. For example, when PHC spikes by 40W within 0.1 seconds, the DRT upper limit automatically drops to 100W to suppress the risk of overshoot.

[0091] The constraint process consists of two steps: Hard Limiting: PHC is compared with SST. If it exceeds 120W, it will force the output to 120W (Upper Clamped Value, UCV). If it is lower than 0W, it will output 0W (Lower Clamped Value, LCV).

[0092] Soft Constraint: Within the SST range, further compare PHC and DRT. For example, if the upper limit of DRT is 100W, and PHC = 105W, then compress it to 100W proportionally (compression factor K_COMP = 0.95). Simultaneously, activate the Integral Separation Mechanism (ISM): When the output is limited, disconnect the integral link (IL) of the PID controller to prevent accumulated errors from causing continuous saturation.

[0093] Before the Safe Heating Command (SHC) is output, it must pass through a Ramp Function Generator (RFG). If the difference between the current SHC and the previous value exceeds a threshold (e.g., 20W), the RFG will gradually increase the output at a preset slope (e.g., 50W / second) to prevent a power surge that could impact the heating element. For example, if the previous SHC was 80W and the newly calculated SHC is 110W, the SHC will gradually increase from 80W to 110W at a rate of 50W / second over 0.6 seconds. All limiting events are recorded in the Status Register (SR) for fault diagnosis.

[0094] The safe heating command is input into the thermal inertia compensation module, and phase correction is performed in combination with the printhead quality parameters to output an optimized heating command. The Thermal Inertia Compensator (TIC) module addresses command delivery delay issues. Printhead Mass Parameters (MP) include: Thermal Capacitance (TC): For example, 380 J / K for a copper alloy heating block; Thermal resistance (TR): For example, the thermal resistance from a ceramic substrate to the environment is 1.2 K / W.

[0095] These parameters are pre-stored in non-volatile memory (NVM). TIC establishes a first-order lag model (FOLM) based on MP: τ = TC × TR (τ is a time constant in seconds). For example, when TC = 380 J / K and TR = 1.2 K / W, τ = 456 seconds, indicating that the command needs to be compensated for 456 seconds in advance to offset thermal inertia.

[0096] Phase correction (PC) is achieved through the Smith Predictor (SP): Establish the prediction model: T_pred(t) = T_real(t) + [SHC(t) - SHC(t-τ)]×(1 - e^(-Δt / τ)). Where T_pred(t) is the predicted temperature at time t, T_real(t) is the actual temperature, and Δt is the sampling interval (1ms).

[0097] The predicted temperature T_pred(t) is fed back to the controller input instead of the actual temperature to compensate for the delay effect in advance.

[0098] For example, when the SHC suddenly increases by 20W, the model immediately calculates the theoretical temperature rise after 456 seconds and increases the equivalent power in advance.

[0099] The generation of the Optimized Heating Command (OHC) incorporates an Adaptive Feedforward (AFF). The system monitors printing speed (PS, mm / s) and material coverage (MC, percentage). When PS > 50 mm / s and MC > 80%, it indicates a significant increase in heat load. In this case, the AFF adds an extra compensation (e.g., +15W), obtained through a look-up table (LUT). Finally, the OHC is filtered by a noise shader (NS) to remove high-frequency components above 10 kHz, preventing resonance in the drive circuit.

[0100] The optimized heating command is converted into a PWM drive signal and output to the printhead heating element to perform temperature control.

[0101] The PWM converter (PWMC) employs triangular wave modulation (TWM). Its core components are: Carrier Generator (CG): Outputs a 100Hz triangular wave (frequency f_CARR=100Hz, peak-to-peak value 0-3V). Comparator (CMP): Scales the OHC command voltage (0-5V) to 0-3V and compares it with a triangular wave; When the OHC voltage is higher than the triangular wave, the output is high; otherwise, the output is low. For example, when OHC = 2.4V, the duty cycle (DC) = 2.4V / 3V × 100% = 80%.

[0102] To adapt to the characteristics of different heating elements, PWMC supports multi-mode switching: Constant Voltage Mode (CVM): Directly outputs a PWM wave, suitable for heating elements with stable resistance; Constant Current Mode (CCM): Adds a Current Feedback Loop (CFL) to adjust the duty cycle in real time via a Hall Sensor (HS) to ensure a constant current, suitable for PTC thermistors; The mode selection is automatically triggered by the Material Type Identifier (MTI). For example, CVM is enabled when printing ABS plastic, and CCM is enabled when printing metal powder.

[0103] The final PWM signal drives the heating element via a power amplification stage (PAS). The PAS includes: Gate Driver (GD): Raises the PWM logic level (0-5V) to the MOSFET drive level (0-12V). H-Bridge Circuit (HBC): Supports bidirectional current control and prevents electrolytic effects from damaging the heating element.

[0104] The system monitors the MOSFET junction temperature (JT) in real time. If it exceeds 150℃, it triggers dynamic frequency derating (FDD) to reduce the PWM carrier frequency from 100Hz to 50Hz to reduce switching losses. The drive status is uploaded to the main control system via the CAN bus (Controller Area Network Bus, CAN) to complete closed-loop control.

[0105] This method intelligently integrates base power and dynamically compensated power, and ensures the safety and feasibility of control commands through an output limiting protection mechanism. The anti-saturation design avoids the risk of control failure under extreme operating conditions. Through power combining and output protection, it fully leverages the advantages of composite control while ensuring the safe and reliable operation of the system, achieving an optimal balance between the accuracy and stability of temperature control in practical applications.

[0106] As can be seen, real-time acquisition of printhead temperature data and decomposition of temperature fluctuations into low-frequency steady-state components and high-frequency transient components using Fast Fourier Transform (FFT) are employed. Based on the printing material type and ambient temperature, the material's thermal response characteristic curve is predicted, generating a thermal response compensation coefficient. A PID controller is used to generate the base heating power for the low-frequency steady-state component, while amplitude feature vectors are extracted from the high-frequency transient components. These feature vectors, along with the thermal response compensation coefficients, are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. Based on the base heating power and dynamic compensation power, a final heating command for the printhead is generated, and the command output range is constrained by an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature. This enables precise dynamic regulation of the printhead temperature, reducing the impact of temperature fluctuations on print quality.

[0107] Another embodiment of the present invention provides a printhead temperature control optimization system, see [link to relevant documentation]. Figure 3 The system may include: The acquisition module 301 is used to acquire printhead temperature data in real time and decompose temperature fluctuations into low-frequency steady-state components and high-frequency transient components through fast Fourier transform. The prediction module 302 is used to predict the thermal response characteristic curve of the material based on the type of printing material and the ambient temperature, and generate a thermal response compensation coefficient. The prediction is based on a mapping relationship library between the material's specific heat capacity, thermal conductivity and ambient temperature. Extraction module 303 is used to generate basic heating power from the low-frequency steady-state component using a PID controller, and simultaneously extract amplitude feature vectors from the high-frequency transient component. The amplitude feature vectors and the thermal response compensation coefficients are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. The generation module 304 is used to generate a final heating command for the printhead based on the basic heating power and the dynamic compensation power, and to constrain the command output range through an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature.

[0108] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0109] Specifically, in this embodiment, the storage medium can be configured to store a computer program for performing the following steps: S201 collects printhead temperature data in real time and decomposes temperature fluctuations into low-frequency steady-state components and high-frequency transient components through fast Fourier transform. S202, based on the type of printing material and the ambient temperature, predict the thermal response characteristic curve of the material and generate a thermal response compensation coefficient, wherein the prediction is based on a mapping relationship library of material specific heat capacity, thermal conductivity and ambient temperature; S203, a PID controller is used to generate the basic heating power for the low-frequency steady-state component, and the amplitude feature vector is extracted from the high-frequency transient component. The amplitude feature vector and the thermal response compensation coefficient are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. S204, based on the basic heating power and the dynamic compensation power, generates a final heating command for the printhead, and constrains the command output range through an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature.

[0110] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0111] Specifically, the aforementioned electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the aforementioned processor, and the input / output device is connected to the aforementioned processor.

[0112] Specifically, in this embodiment, the processor can be configured to perform the following steps via a computer program: S201 collects printhead temperature data in real time and decomposes temperature fluctuations into low-frequency steady-state components and high-frequency transient components through fast Fourier transform. S202, based on the type of printing material and the ambient temperature, predict the thermal response characteristic curve of the material and generate a thermal response compensation coefficient, wherein the prediction is based on a mapping relationship library of material specific heat capacity, thermal conductivity and ambient temperature; S203, a PID controller is used to generate the basic heating power for the low-frequency steady-state component, and the amplitude feature vector is extracted from the high-frequency transient component. The amplitude feature vector and the thermal response compensation coefficient are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. S204, based on the basic heating power and the dynamic compensation power, generates a final heating command for the printhead, and constrains the command output range through an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature.

[0113] The above description, based on the embodiments shown in the figures, details the structure, features, and effects of the present invention. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A method for optimizing printhead temperature control, characterized in that, The method includes: Real-time acquisition of printhead temperature data; temperature fluctuations are decomposed into low-frequency steady-state components and high-frequency transient components using fast Fourier transform. Based on the type of printing material and the ambient temperature, the thermal response characteristic curve of the material is predicted, and a thermal response compensation coefficient is generated. The prediction is based on a mapping relationship library between the material's specific heat capacity, thermal conductivity, and ambient temperature. A PID controller is used to generate the basic heating power for the low-frequency steady-state component, while an amplitude feature vector is extracted from the high-frequency transient component. The amplitude feature vector and the thermal response compensation coefficient are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. Based on the base heating power and the dynamic compensation power, a final heating command for the printhead is generated, and the command output range is constrained by an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature. The real-time acquisition of printhead temperature data is used to decompose temperature fluctuations into low-frequency steady-state components and high-frequency transient components through Fast Fourier Transform, including: Raw temperature data stream is generated by acquiring multiple points on the surface of the printhead using a high-precision thermocouple array at a sampling rate of 1kHz. The raw temperature data stream is input into an adaptive sliding window filter, and the window size is dynamically adjusted according to the noise statistics to output a noise-reduced temperature sequence. The noise-reduced temperature sequence is processed by Fast Fourier Transform, and the Hanning window function is dynamically selected for spectrum analysis to output a full-band temperature spectrum. Based on the full-band temperature spectrum, an energy accumulation algorithm is applied to automatically determine the 0.5Hz segmentation threshold and output the separated low-frequency steady-state component and high-frequency transient component. The step involves predicting the thermal response characteristic curve of the material based on the type of printing material and the ambient temperature, and generating a thermal response compensation coefficient. This prediction is based on a mapping library of the material's specific heat capacity, thermal conductivity, and ambient temperature, and includes: The material identification sensor obtains the printing material type code, and at the same time reads the ambient temperature sensor data, outputting the material type identifier and the real-time ambient temperature value. Input the material type identifier into the material thermal property database, retrieve the corresponding specific heat capacity and thermal conductivity benchmark values, and output the material basic thermal property vector. Input the material's basic thermal property vector and the ambient temperature value into the thermal response prediction model, calculate the temperature compensation offset, and output the preliminary thermal response curve. The initial thermal response curve is input into the real-time calibration module, and the curve parameters are corrected by combining historical temperature fluctuation data to output an optimized thermal response curve. The time constant and gain factor are extracted from the optimized thermal response curve, and the normalization coefficient is calculated by weighted fusion to output the thermal response compensation coefficient.

2. The method according to claim 1, characterized in that, The process involves using a PID controller to generate a base heating power for the low-frequency steady-state component, while simultaneously extracting an amplitude feature vector from the high-frequency transient component. This amplitude feature vector, along with the thermal response compensation coefficient, is input into a pre-trained spiking neural network (SNN) to output dynamic compensation power. The process includes: The low-frequency steady-state component is input into the adaptive PID controller, and the control parameters are automatically tuned according to the temperature change rate, and the basic heating power signal is output. An envelope detection algorithm is applied to the high-frequency transient components to extract peak value, root mean square and zero-crossing rate features, and output amplitude feature vector. The amplitude eigenvector and the thermal response compensation coefficient are concatenated to generate an enhanced eigenvector. The enhanced feature vector is input into a pre-trained spiking neural network model, and the original compensation power is output through spatiotemporal event-driven processing. The original compensation power is input into the exponential smoothing filter, and the smoothing factor is adjusted according to the frequency of temperature fluctuations to output dynamic compensation power.

3. The method according to claim 2, characterized in that, The process of generating a final heating command for the printhead based on the base heating power and the dynamic compensation power, and constraining the command output range through an anti-saturation limiter to achieve dynamic control optimization of the printhead temperature, includes: The basic heating power and the dynamic compensation power are input into the dynamic weighted fusion unit, the weight ratio is adjusted according to the temperature deviation, and the initial heating command is output. The initial heating command is input into the anti-saturation limiter, and the output range is limited by the dual threshold constraint algorithm to output a safe heating command. The safe heating command is input into the thermal inertia compensation module, and phase correction is performed in combination with the printhead quality parameters to output an optimized heating command. The optimized heating command is converted into a PWM drive signal and output to the printhead heating element to perform temperature control.

4. A printhead temperature control optimization system, characterized in that, The system includes: The acquisition module is used to acquire printhead temperature data in real time and decompose temperature fluctuations into low-frequency steady-state components and high-frequency transient components through fast Fourier transform. The prediction module is used to predict the thermal response characteristic curve of the material based on the type of printing material and the ambient temperature, and generate a thermal response compensation coefficient. The prediction is based on a mapping relationship library between the material's specific heat capacity, thermal conductivity and ambient temperature. The extraction module is used to generate a basic heating power from the low-frequency steady-state component using a PID controller, and simultaneously extract the amplitude feature vector from the high-frequency transient component. The amplitude feature vector and the thermal response compensation coefficient are input into a pre-trained pulse neural network (SNN) to output dynamic compensation power. The generation module is used to generate a final heating command for the printhead based on the base heating power and the dynamic compensation power, and to constrain the command output range through an anti-saturation limiter in order to achieve dynamic control optimization of the printhead temperature. The acquisition module is specifically used for: Raw temperature data stream is generated by acquiring multiple points on the surface of the printhead using a high-precision thermocouple array at a sampling rate of 1kHz. The raw temperature data stream is input into an adaptive sliding window filter, and the window size is dynamically adjusted according to the noise statistics to output a noise-reduced temperature sequence. The noise-reduced temperature sequence is processed by Fast Fourier Transform, and the Hanning window function is dynamically selected for spectrum analysis to output a full-band temperature spectrum. Based on the full-band temperature spectrum, an energy accumulation algorithm is applied to automatically determine the 0.5Hz segmentation threshold and output the separated low-frequency steady-state component and high-frequency transient component. The prediction module is specifically used for: The material identification sensor obtains the printing material type code, and at the same time reads the ambient temperature sensor data, outputting the material type identifier and the real-time ambient temperature value. Input the material type identifier into the material thermal property database, retrieve the corresponding specific heat capacity and thermal conductivity benchmark values, and output the material basic thermal property vector. Input the material's basic thermal property vector and the ambient temperature value into the thermal response prediction model, calculate the temperature compensation offset, and output the preliminary thermal response curve. The initial thermal response curve is input into the real-time calibration module, and the curve parameters are corrected by combining historical temperature fluctuation data to output an optimized thermal response curve. The time constant and gain factor are extracted from the optimized thermal response curve, and the normalization coefficient is calculated by weighted fusion to output the thermal response compensation coefficient.

5. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-3 when it is run.

6. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1-3.

Images