A method for controlling color consistency in a bamboo kitchenware carbonization process

Abstract

The application discloses a bamboo and wood kitchen utensil carbonization process color consistency control method, and belongs to the technical field of bamboo and wood processing. The method comprises the following steps: acquiring a multi-dimensional spectrum data vector of a bamboo and wood kitchen utensil surface; determining a multi-dimensional change rate vector of the spectrum data; generating a first candidate control signal and a second candidate control signal based on the spectrum data and the change rate vector; determining a system dynamic state index based on the change rate vector; generating a dynamic weighting coefficient according to the system dynamic state index; and weighting and fusing the first and second candidate control signals by using the dynamic weighting coefficient to generate a final heat regulation control signal. By constructing a double-mode decision fusion composite control system, the application realizes full-working-condition adaptive optimal control, ensures rapid response in the dynamic stage, ensures smooth convergence in the steady state stage, and fundamentally improves the system thermal inertia problem.

Description

Technical Field

[0001] This application relates to the field of bamboo and wood product processing technology, and in particular to an automated control method for the carbonization process of bamboo and wood kitchenware, specifically a method for controlling the color consistency of bamboo and wood kitchenware during the carbonization process. Background Technology

[0002] Bamboo and wood kitchenware, such as cutting boards, bowls, and chopsticks, is widely popular due to its natural and environmentally friendly properties. Carbonization is an important bamboo and wood modification process. By treating bamboo and wood in a high-temperature, oxygen-deficient environment, the sugars, starches, and proteins inside the bamboo and wood can be decomposed, giving the finished product good anti-corrosion and anti-insect properties, and presenting a rustic, uniform dark appearance. However, there are significant technical problems with the control of the carbonization process in existing technologies. Currently, the determination of the carbonization endpoint mostly relies on the visual experience of the operator, who manually adjusts the heating time and power by observing the color change of the bamboo and wood through an observation hole. This method is highly subjective and easily affected by factors such as light, observation angle, and operator fatigue, resulting in obvious color differences between different batches, and even between different locations within the same batch of finished kitchenware, leading to poor product consistency and a low pass rate. To solve this problem, although some solutions have proposed using PID controllers for closed-loop control, the carbonization kettle, as a large thermodynamic system, has huge physical thermal inertia, resulting in a serious lag in control response. Simple feedback control is prone to temperature overshoot, meaning the system only starts reducing heat when the color reaches the target value. However, the significant thermal inertia continues to "push" the charring reaction, resulting in an excessively dark final color that is difficult to precisely control at the target point. Therefore, overcoming system thermal inertia and achieving automated, high-precision control of charring color is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0003] The purpose of this application is to provide a method for controlling the color consistency of bamboo and wood kitchenware during the carbonization process, aiming to solve the technical problems in the prior art where the control of carbonization color relies on manual experience, has a low degree of automation, and is difficult to overcome the large color difference of the finished product caused by the thermal inertia of the system.

[0004] To achieve the above objectives, in a first aspect, this application provides a method for controlling the color consistency during the carbonization process of bamboo and wood kitchenware, comprising: acquiring multidimensional spectral data vectors of the surface of bamboo and wood kitchenware in real time; determining a multidimensional rate of change vector of the current spectral data based on historical spectral data; generating at least two candidate control signals in parallel, the at least two candidate control signals comprising: generating a first candidate control signal based on the current deviation vector between the spectral data vector and a target spectral vector, and the rate of change vector; predicting a multidimensional spectral vector at a future time based on the current spectral data vector, the rate of change vector, and a preset prediction time constant, and generating a second candidate control signal based on the prediction deviation vector between the multidimensional spectral vector at the future time and the target spectral vector; determining a system dynamic state index based on the multidimensional rate of change vector; and dynamically weighting and fusing the first candidate control signal and the second candidate control signal according to the system dynamic state index to generate a final heat regulation control signal.

[0005] The core innovation of the technical solution provided in this application lies in the construction of a dynamic dual-modal decision fusion composite control system. This system no longer relies on any single control strategy, but instead operates two control models with complementary advantages in parallel: one is a fast response model based on direct compensation of the current state, and the other is a smooth control model based on explicit prediction of the physical model. More importantly, this application introduces a decision fusion arbitrator, which intelligently and dynamically adjusts the weights of the two control models in the final decision by monitoring the system's dynamic state (i.e., the intensity of the carbonization reaction) in real time. During the dynamic phase of the carbonization process, the system automatically emphasizes the fast response model to ensure rapid response to temperature changes; in the steady-state phase approaching the target, the system smoothly switches to emphasizing the smooth prediction model to achieve precise convergence without overshoot. This full-condition adaptive optimal control strategy achieves unexpected technical effects that no single control model can reach, combining both speed and stability, fundamentally solving the control problem of thermal inertial systems.

[0006] Optionally, determining a system dynamic state index based on the multidimensional rate of change vector includes: calculating the norm of the multidimensional spectral data vector as the system dynamic state index.

[0007] Optionally, the step of dynamically weighting and fusing the first candidate control signal and the second candidate control signal according to the system dynamic state index includes: calculating a dynamic weighting coefficient through a preset transition function based on the comparison relationship between the system dynamic state index and a preset dynamic threshold; and performing a weighted summation of the first candidate control signal and the second candidate control signal based on the dynamic weighting coefficient to obtain the final heat regulation control signal.

[0008] Optionally, the transition function is a Sigmoid function.

[0009] Optionally, when the system dynamic state index is greater than the dynamic threshold, the dynamic weighting coefficient is increased to increase the weight of the first candidate control signal in the final heat regulation control signal; when the system dynamic state index is less than the dynamic threshold, the dynamic weighting coefficient is decreased to increase the weight of the second candidate control signal in the final heat regulation control signal.

[0010] Optionally, generating a first candidate control signal includes: obtaining a first proportional combination of the current deviation vector to obtain a feedback control component; obtaining a feedforward combination of the rate of change vector to obtain a feedforward compensation component; and combining the feedback control component and the feedforward compensation component to generate the first candidate control signal.

[0011] Optionally, predicting a multidimensional spectral vector for a future moment includes: combining the rate of change vector with the prediction time constant to obtain a prediction increment vector; and combining the current spectral data vector with the prediction increment vector to obtain the multidimensional spectral vector for the future moment.

[0012] Optionally, the multidimensional spectral data vector is the spectral energy distribution vector of the bamboo and wood kitchenware surface in the near-infrared band.

[0013] Optionally, the predicted time constant is pre-calibrated based on the physical thermal inertia of the carbonization vessel.

[0014] Optionally, the method further includes: converting the final heat regulation control signal into an adjustment command for the duty cycle of the pulse width modulation signal of the carbonization kettle heating unit, so as to control the energy output of the heating unit. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a method for controlling the color consistency of bamboo and wood kitchenware carbonization process according to an embodiment of this application, which shows the steps of generating two candidate control signals in parallel (S310 and S320).

[0017] Figure 2 It is implemented according to one embodiment of this application and is used for execution. Figure 1 The system functional block diagram of the control method shows the system architecture, including the parallel control signal generation module and the decision fusion arbitration module, as well as the data flow relationships between the modules.

[0018] Figure 3 This is a schematic diagram comparing the output characteristics of the first candidate control signal (fast response type) and the second candidate control signal (smooth prediction type) during an exemplary dynamic change process of a system, used to illustrate the complementarity of the two control strategies.

[0019] Figure 4 This is a schematic diagram illustrating the functional relationship between the dynamic weighting coefficients and the dynamic state index of the system, used to explain the decision fusion logic in this application. It shows the smooth transition mechanism of the control weights between the dynamic and steady-state stages.

[0020] Figure 5 This is a schematic diagram comparing the system response curve of the fusion control method proposed in this application with the response curve of the traditional control method, which is used to demonstrate the effect of this application in suppressing control overshoot and achieving fast and smooth convergence. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0023] Method implementation examples refer to Figure 1 One embodiment of this application provides a method for controlling the color consistency during the carbonization process of bamboo and wood kitchenware. In a specific implementation, this method is executed by a control system consisting of an infrared spectrophotometer, a carbonization kettle heat regulating valve, and a hard logic processing chip. The technical solution provided in this application constructs a dynamic dual-modal decision fusion composite control system, thereby eliminating reliance on any single fixed control strategy. Instead, it can intelligently and smoothly switch and fuse between two complementary control modes based on the real-time dynamic characteristics of the carbonization process.

[0024] The first control mode is a fast response control mode based on direct compensation of the current state. Its advantage is that it has a very high following ability to system disturbances and rapid changes. The second mode is a smooth control mode based on physical model prediction. Its advantage is that it has a forward-looking view of future states and can achieve accurate convergence without overshoot when approaching the target.

[0025] This method introduces a decision fusion arbitrator to assess the intensity of the carbonization reaction in real time. Based on this assessment, it dynamically allocates the weights of the two control modes in the final decision, thereby achieving optimal control that combines speed and stability across the entire carbonization process. This comprehensive performance is unattainable by a single control model. This method solves the technical problems of existing technologies that rely on human experience and cannot overcome system thermal inertia, resulting in poor color consistency of the finished product. It achieves the beneficial effect of stable and automated control of the color at the carbonization endpoint.

[0026] Before detailing the method and process proposed in this application, we will first define some key technical terms in this application.

[0027] In the context of this application, a multidimensional spectral data vector refers to an ordered set of values ​​constructed to characterize the reflectivity of bamboo and wood kitchenware surfaces to electromagnetic waves of different wavelengths at a specific moment. Each dimension (or component) of this vector corresponds to a specific narrow wavelength range, and the value in that dimension quantifies the intensity of reflected energy detected within that wavelength range. Therefore, this vector is not a single numerical value, but rather constitutes a high-dimensional spectral fingerprint whose shape, amplitude, and time-varying characteristics can map the dynamic pyrolysis and chemical bond changes of the internal chemical components of bamboo and wood, such as hemicellulose, cellulose, and lignin, during the carbonization process.

[0028] A multidimensional rate of change vector refers to a vector with the same dimensions as the multidimensional spectral data vector, designed to quantify the speed and direction of the evolution of the spectral fingerprint over time. Each component of this vector represents the first derivative of the reflected energy intensity at the corresponding wavelength with respect to time, i.e., the rate of change. Therefore, this vector as a whole describes the velocity of the spectral state in multidimensional space, its direction indicates the main trend of state evolution, and its norm (or modulus) quantifies the intensity of the evolution, i.e., the macroscopic rate of the carbonization reaction.

[0029] Prediction time constant ( The term "time parameter" refers to a scalar parameter with the dimension of time. Its physical meaning is an engineered estimate of the inherent response delay of the controlled system, namely the thermodynamic system of the carbonization kettle. It quantifies the time required from applying an action at the control input (e.g., adjusting the power of the heating unit) to the system output (e.g., a measurable change in the spectrum of the bamboo or wood surface) exhibiting the corresponding result.

[0030] S100: Real-time acquisition of multidimensional spectral data vectors of bamboo and wood kitchenware surfaces.

[0031] In this step, the sensing front end of the control system, namely the infrared spectrophotometer module, performs continuous non-contact optical detection on the surface of bamboo and wood kitchenware in the high-temperature, low-oxygen, or vacuum environment of the carbonization kettle. The main chemical components of bamboo and wood include cellulose, hemicellulose, and lignin. During carbonization, the relatively unstable hemicellulose undergoes pyrolysis first, resulting in the breakage of its macromolecular chains and changes in functional groups (such as hydroxyl groups -OH and carbonyl groups C=O). The vibrational and absorption characteristics of these functional groups show a significant response in the near-infrared spectral region. In this application, the spectral information is not compressed into a single index, but rather treated as a whole, i.e., a multi-dimensional vector, thereby preserving the complete spectral fingerprint reflecting the chemical state of the bamboo and wood surface to the greatest extent possible.

[0032] In a preferred embodiment, the spectral detection range is limited to the near-infrared band of 900 nm to 1700 nm. This band can effectively avoid ambient light interference in the visible light range and maximize the capture of chemical bond information related to the degree of carbonization, such as the 1100-1200 nm band related to CH bonds in lignin and the 1400-1500 nm band related to OH bonds in cellulose and hemicellulose.

[0033] In the specific execution process, the light emitted from the broadband light source (e.g., a 50-watt halogen lamp) inside the infrared spectrophotometer module passes through an optical collimation and focusing system made of quartz glass (e.g., including a collimating lens and a focusing lens) to form a detection spot with a diameter of approximately 10 mm, which is perpendicularly irradiated onto the surface of the bamboo or wood kitchenware to be tested. The light diffusely reflected back from the surface is collected by a receiving optical system arranged at a specific angle (e.g., 45 degrees) to the irradiated light path to minimize interference from specular reflection. The collected light is guided through an optical fiber to a beam splitting unit (e.g., a diffraction grating using a planar holographic grating with a line density of 300 lines / mm). The diffraction grating disperses the reflected light according to different wavelengths, forming a spectrum, which is projected onto a linearly arranged detector array (e.g., an InGaAs sensor array containing 256 pixel units). Each pixel in the array corresponds to a specific narrowband wavelength. The photons it receives are converted into electrical charges, which are then integrated, amplified, and converted from analog to digital by the back-end readout circuit. Finally, the signal processing circuit inside the module integrates the output signal of the entire array after preprocessing such as dark current correction and gain correction to form a discretized multidimensional spectral data vector, denoted as . Each dimension of this vector corresponds to a specific wavelength, and its value is the spectral energy density at that wavelength. This process is performed periodically with extremely high time resolution, for example, every 100 milliseconds (...). This generates a new spectral data vector, thus providing high-density time-series data for subsequent dynamic predictions.

[0034] For example, to illustrate the technical concept of multidimensional spectral data vectors, assume that the infrared spectrophotometer module discretizes the 900 nm to 1700 nm wavelength band into 81 equally spaced sampling points, i.e., a wavelength resolution of 10 nm. Then, at any sampling time... The acquired real-time spectral data is a vector containing 81 elements. Each element in the vector Represents wavelength The spectral energy density value measured at [location]. In a specific numerical example, at a certain intermediate moment in the carbonization process. The obtained spectral data vector can be The values ​​here can have a specific physical unit, such as watts per square meter per steradian per nanometer. For ease of calculation and understanding, in subsequent examples, it can be normalized or treated as a dimensionless relative intensity unit (au). This vector fully depicts the optical fingerprint of the bamboo surface at that moment, serving as the data source for all subsequent calculations and decisions. Each dimension of this vector carries independent information about the carbonization state, and the evolution of the entire vector as a whole, in terms of its shape and value, can map the dynamic evolution of the internal chemical structure of the bamboo. This method of preserving full-dimensional information allows the control system to perceive more subtle state changes, thus laying the foundation for predicting future states.

[0035] S200: Based on historical spectral data, determine the multidimensional rate of change vector of the current spectral data.

[0036] This step quantifies the rate and direction of change of the current spectral state vector. This step is executed by the dynamic state calculation module within the hard logic processing chip. This module internally maintains a data cache structure, typically a fixed-length First-In-First-Out (FIFO) queue, to store the spectral data vectors from the most recent N sampling times. For example, it can be set... The queue always stores the data from the current moment. to the past moment The system receives 10 consecutive spectral vectors. The length N of the FIFO queue is a parameter that can be adjusted according to the actual application scenario, and can be exemplarily taken as 2 to 10. A smaller N value (e.g., N=2, i.e., only using the current and previous time-series data) makes the system more sensitive to the rate of spectral change, but also more sensitive to measurement noise. Conversely, a larger N value allows for the use of more complex numerical differentiation methods (e.g., slope calculation based on multi-point fitting), thereby enabling effective smoothing and filtering of the input signal and enhancing the system's ability to suppress high-frequency noise. Those skilled in the art can make trade-offs between sensitivity and stability based on the actual signal-to-noise ratio of the measured signal and the requirements for system response speed. When new spectral data... Upon arrival, it is pushed to the head of the queue, while the oldest data at the tail of the queue is discarded.

[0037] Based on this queue storing historical data, the module uses numerical differentiation to estimate the rate of change vector of the spectral data vector. In some implementations, a first-order backward difference method can be used. This method uses only the two most recent data points in the queue, i.e., the current time step. Data and the previous sampling time Data We use this to calculate the rate. Its mathematical expression is: .in, This is the sampling time interval (e.g., 100 ms). This calculation is performed element-wise, meaning that the difference operation is performed independently for each dimension component of the vector. Therefore, the resulting rate of change... It is also a multidimensional vector with the same dimensions as the spectral data. Each component represents the rate of change of spectral energy over time at the corresponding wavelength. Because the bamboo and wood darken in color and their surface reflectivity decreases during carbonization, therefore... The value of is monotonically decreasing, thus leading to The components of this vector are typically negative. This vector not only describes the rate of change of the overall color, but also depicts the pattern of change, i.e., in which bands the change is faster, providing important dynamic state information for subsequent decision fusion arbitration.

[0038] Alternatively, to suppress measurement noise while calculating the rate of change, a more robust numerical differentiation method can be employed. For example, the central difference method can be used, which uses data from two time points before and after the current time for calculation: This method offers higher accuracy but introduces a delay of one sampling period. In another, more preferred embodiment, a Savitzky-Golay filter can be used. This method involves polynomial fitting of the data points within a sliding time window and then analytically calculating the derivative of the polynomial at that point. For example, a second-order polynomial Savitzky-Golay filter with a window length of 5 can be chosen, which provides a more accurate rate-of-change estimate while effectively smoothing noise, without introducing excessive phase delay.

[0039] For example, to clearly illustrate the calculation process of the multidimensional rate of change vector, it is assumed that only two characteristic wavelengths are monitored, i.e., all vectors are two-dimensional vectors, and a first-order backward difference method is used. The sampling interval of the system is assumed to be... It is 1.0 second (s).

[0040] exist At 1 second, the measured spectral vector is (au).

[0041] exist At 1 second, the newly measured spectral vector is (au).

[0042] The dynamic state calculation module will perform the following element-wise calculation to determine the rate of change vector. :

[0043] Rate of change in the first dimension: (au / s)

[0044] The rate of change in the second dimension: (au / s)

[0045] Therefore, in The multidimensional rate of change vector at time 1 second is (au / s). This vector precisely tells the control system that the current spectral state of the bamboo surface is evolving along a specific direction (in two-dimensional space, from point (65, 48) to point (63, 45)) at a specific speed. This vector will be broadcast to the two subsequent parallel control channels and the decision fusion arbitrator.

[0046] S300: Generates candidate control signals in parallel under at least two modes.

[0047] The system will acquire real-time status information ( and The signal is distributed to two parallel controllers to generate two candidate control signals simultaneously. One controller is biased towards handling contingencies, while the other is biased towards planning. This parallel processing architecture ensures that the system can hold two response strategies at any given time, providing a basis for subsequent intelligent decision-making.

[0048] S310: Generate the first candidate control signal based on direct compensation of the current state.

[0049] This step is performed by the indirect compensation controller (which can be considered "Expert A") in the parallel control signal generation module. This controller is used to react quickly to current state deviations and dynamic trends. Instead of building an explicit model of the future, it directly counteracts the dynamic trends of the system through a feedforward compensation term, thereby achieving rapid following.

[0050] The specific calculation process includes two parts. First, calculate the current deviation vector. ,in It is the preset target spectral vector. Then, a feedback control component is generated based on this current deviation vector. ,in This is the proportional gain coefficient of the control channel. Secondly, based on the rate of change vector... Generate a feedforward compensation component ,in This is the feedforward gain coefficient. Finally, these two components are combined to generate the first candidate control signal. This signal is characterized by occurring when the carbonization reaction is vigorous (i.e., When the absolute value of is large, the feedforward compensation term will generate a strong braking effect to prevent the system from losing control due to over-response.

[0051] In this step, the proportional gain coefficient and feedforward gain coefficient These are all key parameters that need to be pre-tuned. It primarily affects the system's response strength to static errors, and is dimensionless. This determines the system's ability to suppress dynamics, measured in seconds (s). The tuning of these two parameters requires comprehensive consideration to achieve a balance between response speed and disturbance rejection.

[0052] Continuing with the two-dimensional vector example, let's define the target spectral vector. And set the control parameters for this channel as follows: (dimensionless) s. In Seconds, known and .

[0053] Calculate the current deviation vector: .

[0054] Calculate the feedback component: .

[0055] Calculate the feedforward component: .

[0056] Calculate the first candidate control signal: .

[0057] S320: Generate a second candidate control signal based on explicit predictions of future states.

[0058] This step is performed in parallel with S310 by an explicit predictive controller (which can be considered "Expert B"). This controller is used to predict the state of the system a short time window in the future and, based on this prediction, to formulate a smoother and more forward-looking control strategy.

[0059] The controller has a key preset parameter—the prediction time constant. This parameter represents an engineering estimate of the system's thermal inertia delay, measured in time, and can be pre-calibrated experimentally. The calibration process may include the following steps: First, stabilize the carbonization vessel at an initial temperature and continuously record spectral data; then, apply a known power step change to the heating unit (e.g., instantaneously increasing the duty cycle from 20% to 50%); next, continuously monitor the change curve of the spectral vector and record the time elapsed from the application of the step until each component of the spectral vector reaches 63.2% of its new steady-state value; finally, average the times measured in multiple experiments to obtain a reliable estimate. Value. For example, a medium-sized carbonization autoclave... The value may be between 4.0 seconds and 8.0 seconds.

[0060] The controller performs a vectorized linear extrapolation to predict the future state, and its mathematical expression is: In obtaining the predicted future spectral vector Then, the controller calculates its relationship with the target spectral vector. The prediction deviation vector between Finally, a second candidate control signal is generated based on this prediction deviation vector. ,in This is the proportional gain coefficient of the control channel. This signal is designed to eliminate future deviations, thus smoothly guiding the system to the target point, especially in the steady-state phase near the target, effectively avoiding oscillations.

[0061] For example, continuing with the two-dimensional vector example. The control parameters for this channel are set as follows. s, (Dimensionless). In Seconds:

[0062] Predict future spectral vectors: .

[0063] Calculate the prediction bias vector: .

[0064] Calculate the second candidate control signal: .

[0065] S400: Based on the dynamic state of the system, candidate control signals are dynamically weighted and fused.

[0066] This step is performed by the decision fusion arbitration module. This module receives candidate control signals from two parallel controllers. and However, instead of simply averaging or choosing one, the weights of the two are intelligently determined based on a real-time assessment of the intensity of the current carbonization process, ultimately generating a final control signal that combines the advantages of both.

[0067] S410: Determine the dynamic status indicators of the system.

[0068] To quantify the intensity of the carbonization process, the arbitration module requires an objective metric. In this embodiment, the norm (i.e., its geometric length or modulus) of the multidimensional rate of change vector is selected, denoted as [missing numerator]. , as a dynamic state indicator of the system. If Then its norm calculation formula is: The magnitude of this scalar value intuitively reflects the speed at which the spectral fingerprint moves in the multidimensional state space. A large value indicates a vigorous carbonization reaction and that the system is in a dynamic change phase. A smaller value indicates a slower response and a more stable system.

[0069] Alternatively, to reduce computational complexity, especially in resource-constrained hard logic chips, system dynamic state indicators can be obtained using other computationally simpler methods. For example, the Manhattan norm (L1 norm) can be used, i.e. It replaces the operations of calculating square roots and squares with simple summation of absolute values. Alternatively, the infinity norm (L∞ norm) can be used, i.e. It only needs to compare and identify the rate of change of the dimension in which the change is most dramatic. These alternative indicators can also effectively reflect the dynamics of the system.

[0070] For example, in Seconds, known Using the standard Euclidean norm (L2 norm):

[0071] Calculate the dynamic state indicators of the system: (au / s).

[0072] S420: Calculate the dynamic weighting coefficients.

[0073] The arbitration module calculates the system dynamic status indicators based on S410. To calculate a dynamic weighting coefficient The value of this coefficient is limited to the range of [0, 1]. Used to determine the first candidate control signal (Fast response type) weights. To achieve a smooth transition from one control mode to another and avoid system jitter caused by hard switching, a preset transition function is used. In a preferred embodiment, this transition function is the Sigmoid function. Its mathematical expression is: .in, It is a preset dynamic threshold used to distinguish between the dynamic stage and the steady-state stage; This is the slope parameter of the Sigmoid function, used to control the width of the transition band (i.e., the abruptness of the transition). When Much larger hour, Approaching 1; when much smaller hour, Approaching 0.

[0074] The dynamic threshold and slope parameters These are key parameters that need to be calibrated according to specific process requirements. The setting determines the boundary between the system's perceived dynamic and steady-state states. For example, through experimental observation, it can be determined that when the carbonization reaction rate is below a certain value, the system's overshoot risk is significantly reduced; this value can be set as... . The value determines the smoothness of mode switching; a larger value results in a smoother transition. The value makes the switching process more like a switch, in Completed quickly nearby; smaller The value makes the transition region wider, and the fusion effect of the two modes lasts longer.

[0075] The dynamic threshold and slope parameters Calibration can be performed according to specific process requirements. In an exemplary calibration process, multiple complete carbonization experiments can be conducted first, and the entire process can be recorded. Curve. By analyzing the curve, we can find the inflection point region where the system transitions from a rapid dynamic phase to a smooth steady-state phase. This region corresponds to... The median of the numerical range can be used as The initial setting value. Then, it can be fixed. Adjust the slope parameter Larger The smaller value makes the mode switching process faster, suitable for scenarios requiring high purity of control mode; This value results in a wider transition region and a longer fusion time between the two modes, making it suitable for scenarios requiring extreme smoothness. Those skilled in the art can observe the different effects through several iterative experiments. The optimal value is determined by measuring the smoothness of the system's response in the transition region.

[0076] For example, setting a dynamic threshold au / s, slope parameter .exist Seconds, known au / s.

[0077] Calculate the dynamic weighting coefficients: .

[0078] This coefficient value, close to 1, indicates that in this phase of rapid and dynamic change, system decisions should primarily rely on the fast-response controller A.

[0079] S430: Performs dynamic weighted fusion to generate the final control signal.

[0080] The arbitration module uses the dynamic weighting coefficients calculated by S420. For two candidate control signals and A weighted sum is performed to generate the final heat regulation control signal. Its fusion formula is: In this formula, when When the value approaches 1 (dynamic stage), the final signal is mainly composed of Contribution; when When it approaches 0 (steady-state stage), the final signal is mainly composed of Contribution. During the transition phase, it achieves smooth superposition of the two control signals, combining the advantages of both.

[0081] For example, in At the second (dynamic phase), it is known , as well as .

[0082] Calculate the final control signal:

[0083] .

[0084] To further demonstrate the superiority of the fusion strategy proposed in this application, an example of a steady-state phase is provided. Assume that in... At 10:00, the system was very close to the target and measured... , .

[0085] S200: .

[0086] S310: , .

[0087] S320: , , .

[0088] S410: au / s.

[0089] S420: .

[0090] S430: .

[0091] By comparing the calculations in the two stages, it can be seen that in the dynamic stage, the final decision ( It is closer to a fast-response A controller ( In the steady state phase, the final decision ( The B controller, which is almost entirely responsible for smooth predictions, The above example demonstrates how the dynamic decision fusion method proposed in this application achieves adaptive optimal control across all operating conditions.

[0092] S500: Executes the final thermal regulation control signal.

[0093] This step is performed by the heat regulation execution module. This module receives the final control signal vector generated by the decision fusion arbitration module. Since the signal is a multidimensional vector and its numerical range is not normalized and may contain negative values, it needs to be mapped and limited to convert it into a duty cycle instruction within the range of [0, 100%] that can be directly used by physical actuators (such as PWM generators).

[0094] For example, S500 includes:

[0095] S510: Perform vector to scalar conversion.

[0096] This sub-step aims to incorporate multidimensional control intentions. Dimensionality reduction, condensing into a single scalar value that can represent the overall control strength and direction. .because The components may have different signs and magnitudes, and simple processing methods may lose critical control information. Therefore, this application discloses at least three optional conversion implementation methods with different technical focuses, which can be selected by those skilled in the art according to specific hardware resources and control accuracy requirements.

[0097] The first possible implementation method: weighted average method.

[0098] In this method, scalar control value By adjusting the final control signal vector It is obtained by weighted summation of the components in each dimension. Its mathematical expression is: ,in It is the dimension of the vector. It is the component of the vector in the i-th dimension, and These are preset weights corresponding to the i-th dimension. The weights... The setting is not arbitrary, but based on prior knowledge of the physical characteristics of the system.

[0099] In one embodiment, weight It can be set to correspond to the wavelength The sensitivity or signal-to-noise ratio is directly proportional to the degree of carbonization. For example, the correlation between reflectivity at different wavelengths and the color difference of the final product can be analyzed through preliminary experiments, and the band with the strongest correlation can be assigned a higher weight. For instance, if it is found that the signal in the 1450nm band most stably reflects the carbonization endpoint, then the 1450nm band can be set as the most suitable wavelength for carbonization. The weights of other bands are reduced accordingly, and the sum of all weights The advantage of this approach is that it can highlight the control effect of key characteristic bands.

[0100] The second optional implementation method is the norm symbol mixing method.

[0101] In this method, the conversion process consists of two steps. The first step is to calculate the control signal vector. norm (e.g., L2 norm) The first step is to obtain the overall "strength" or "energy" of the control action, which is a scalar value without direction. The second step is to assign a sign (positive or negative) to this strength value that represents the overall control direction. In one embodiment, the sign can be determined by the sign of the component with the largest absolute value in the vector. For example, if... The component with the largest absolute value is -8.867, so the final scalar control value is... .if The component with the largest absolute value is 5.0, so the final scalar control value is... The advantage of this approach is that it ensures the intensity of the control action is proportional to the magnitude of the deviation in multidimensional space, while also reflecting the direction of the most urgent control needs.

[0102] The third alternative implementation method is the maximum absolute value method.

[0103] In this method, select directly The component with the largest absolute value is taken as the final scalar control value. Its mathematical expression is: where index k satisfies For all This strategy is valid. It assumes that the most drastic deviation component represents the most significant control contradiction, and therefore allocates all control resources to responding to this contradiction. This method is computationally simple, has the lowest hardware implementation cost, and is particularly suitable for situations requiring the fastest possible response to specific mutations.

[0104] S520: Perform scalar to duty cycle mapping.

[0105] This sub-step is responsible for converting the scalar control values ​​generated by S510, which have positive and negative signs and arbitrary numerical ranges. This is converted into a standard pulse width modulation (PWM) duty cycle command with a range between [0%, 100%]. This process is implemented using a pre-calibrated linear mapping function combined with saturation truncation logic.

[0106] In one specific embodiment, the linear mapping function takes the form of: .in, It is the intercept, representing the base duty cycle at "zero control input"; The slope represents the control gain. The calibration process for these two parameters is as follows:

[0107] Determine the baseline duty cycle First, without applying any control (i.e. In this case, a PWM duty cycle is determined that can maintain the carbonization vessel at a stable temperature near the target process temperature (e.g., 180°C) without rising or falling. This is typically determined experimentally and represents the energy input required to overcome the system's natural heat dissipation. For example, this reference duty cycle can be determined as follows: .

[0108] Determine the control gain Secondly, determine a typical, desired maximum effective control value. And determine the corresponding expected maximum heating duty cycle. (A certain margin is usually left, such as 95%). For example, by analyzing historical data or simulations, it is determined that when... At +10 au, the system should heat at near full power. Therefore, a second mapping point can be established. ,Right now Utilizing these two points and The slope can be calculated. (% / au). Therefore, the complete linear mapping function is... This function also applies to negative control inputs, such as a negative . This value will result in a duty cycle of less than 25%, achieving the effect of cooling or reducing heating.

[0109] The initial duty cycle value was calculated using a linear function. Then, it must be subjected to saturation truncation to ensure that the final instruction is physically feasible. The logic is as follows:

[0110] like The final duty cycle instruction .

[0111] like The final duty cycle instruction .

[0112] otherwise, .

[0113] S530: Execute duty cycle instructions.

[0114] Finally, the final duty cycle command generated by S520 is within the range of [0, 100%]. The signal is fed into the PWM signal generator of the thermal regulation execution module. This generator (usually a built-in peripheral of a hard logic processing chip) will, according to... The value of the duty cycle is used to generate a square wave signal with a corresponding high-low time ratio. This square wave signal then drives a high-power switching element (such as a MOSFET or IGBT), which in turn switches the main circuit current flowing to the carbonization kettle heating unit (such as a resistance heating wire) at an extremely high frequency. In this way, the average power obtained by the heating unit macroscopically is related to the duty cycle. This is directly proportional to the energy input of the carbonization process, thus enabling control over the energy input.

[0115] System Implementation Examples

[0116] Reference Figure 2 To perform the above method, this application also provides a control system. This system can be integrated into a bamboo and wood kitchenware carbonization device, and includes:

[0117] A spectral information acquisition module 100 is configured to execute S100, which involves acquiring multidimensional spectral data vectors from the surface of bamboo and wood kitchenware in real time. In its specific implementation, this module is an integrated optical probe, including a broadband infrared light source, collimating and focusing optical elements, a beam splitter, and an InGaAs sensor array. It performs periodic acquisition via a hardware trigger signal and sends the raw sensor readings to the main processing chip via a high-speed serial interface (such as SPI).

[0118] A dynamic state calculation module 200, communicatively connected to the spectral information acquisition module 100, is configured to execute S200, i.e., determine the multidimensional rate of change vector of the current spectral data. In a hard logic implementation, this module may consist of a set of digital signal processing (DSP) logic units, including a block random access memory (BRAM) for storing historical data, and an arithmetic logic unit (ALU) for performing vector subtraction and division operations.

[0119] A parallel control signal generation module 300, connected to the dynamic state calculation module 200, internally includes: an indirect compensation controller 310 configured to execute S310 to generate a first candidate control signal; and an explicit prediction controller 320 configured to execute S320 to generate a second candidate control signal. In the FPGA implementation, these two controllers can be designed as two independent, parallel-running state machines or data flow processing pipelines, sharing state information input from upstream modules.

[0120] A decision fusion arbitration module 400, whose inputs are connected to the dynamic state calculation module 200, the indirect compensation controller 310, and the explicit predictive controller 320 respectively, is configured to execute S400, that is, to dynamically weight and fuse the first and second candidate control signals to generate the final heat regulation control signal. This module is the core of the hardware implementation and may include a dedicated hardware multiplier and a CORDIC (Coordinate Rotation Digital Computer) core for calculating the square root and square operations of the vector norm, as well as a look-up table (LUT) or piecewise linear approximation circuit for implementing the sigmoid function.

[0121] A heat regulation execution module 500, connected to the decision fusion arbitration module 400, is configured to execute S500, i.e., receive the final heat regulation control signal and accordingly precisely control the energy output of the heating unit of the carbonization kettle. This module typically consists of a standard PWM signal generator peripheral and an external high-power drive circuit (such as a MOSFET or IGBT driver board).

[0122] In a specific hardware implementation, the functions of the dynamic state calculation module 200, the parallel control signal generation module 300, and the decision fusion arbitration module 400 can all be integrated onto a single hard logic processing chip, such as a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC). This hardware-level implementation ensures that the execution of all core algorithms has extremely low latency and high determinism.

[0123] It is worth noting that the dual-modal fusion logic proposed in this application differs from the conventional applications of some traditional control theories in certain aspects. During the intensely dynamic phases of the carbonization process (e.g., the rapid temperature rise period, corresponding to a high rate-of-change vector norm), explicit predictions based on a fixed physical model (i.e., the strategy employed by the second candidate control signal) often produce non-negligible prediction errors due to the drastic nonlinear drift of model parameters (such as thermal conductivity and activation energy) with temperature and material state. In this case, control relying on such predictions may actually lead to instability. In contrast, direct compensation control based on the current deviation and rate of change (i.e., the strategy employed by the first candidate control signal), although simpler in model, offers the most direct and rapid response to the current state, exhibiting stronger robustness in dynamic tracking. Furthermore, in the steady-state phase approaching the target, system parameters tend to stabilize. At this point, the accuracy of the explicit prediction model is greatly improved, and its foresight effectively eliminates the small overshoot caused by integral effects, achieving smoother static error convergence than direct compensation control. Therefore, the fusion logic adopted in this application, which focuses on direct compensation in the dynamic stage and explicit prediction in the steady-state stage, is aimed at achieving optimal control across the entire operating range for the specific physicochemical process of bamboo and wood carbonization.

[0124] Those skilled in the art will understand that the systems, apparatuses, modules, and methods disclosed in the above embodiments can be implemented in other ways. For example, the description of the above apparatus embodiments is merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0125] Furthermore, the functional modules described in the several embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0126] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by a program instructing related hardware. The aforementioned program can be stored in a computer-readable storage medium, and when executed, the program performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as a portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0127] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for controlling the color consistency during the carbonization process of bamboo and wood kitchenware, characterized in that, include: Obtain multidimensional spectral data vectors of bamboo and wood kitchenware surfaces; Based on historical spectral data, determine the multidimensional rate of change vector of the current spectral data; At least two candidate control signals are generated in parallel, the at least two candidate control signals including: Based on the current deviation vector between the spectral data vector and a target spectral vector, and the rate of change vector, a first candidate control signal is generated. Based on the current spectral data vector, the rate of change vector, and a preset prediction time constant, a multidimensional spectral vector at a future time is predicted, and a second candidate control signal is generated based on the prediction deviation vector between the multidimensional spectral vector at the future time and the target spectral vector. Based on the multidimensional rate of change vector, a system dynamic state index is determined; Based on the system dynamic state index, the first candidate control signal and the second candidate control signal are dynamically weighted and fused to generate a final heat regulation control signal. The determination of a system dynamic state index based on the multidimensional rate of change vector includes: Calculate the norm of the multidimensional rate of change vector to serve as an index of the system's dynamic state. The step of dynamically weighting and fusing the first candidate control signal and the second candidate control signal based on the system dynamic state index includes: Based on the comparison relationship between the system dynamic state index and a preset dynamic threshold, a dynamic weighting coefficient is calculated through a preset transition function. Based on the dynamic weighting coefficients, the first candidate control signal and the second candidate control signal are weighted and summed to obtain the final heat regulation control signal.

2. The method according to claim 1, characterized in that, The transition function is a Sigmoid function.

3. The method according to claim 2, characterized in that, When the system dynamic state index is greater than the dynamic threshold, the dynamic weighting coefficient increases to increase the weight of the first candidate control signal in the final heat regulation control signal; when the system dynamic state index is less than the dynamic threshold, the dynamic weighting coefficient decreases to increase the weight of the second candidate control signal in the final heat regulation control signal.

4. The method according to claim 1, characterized in that, The generation of a first candidate control signal includes: Obtain a first proportional combination of the current deviation vector to obtain a feedback control component; A feedforward combination of the rate of change vector is obtained to obtain a feedforward compensation component; The feedback control component is combined with the feedforward compensation component to generate the first candidate control signal.

5. The method according to claim 1, characterized in that, The multidimensional spectral vector for predicting a future time includes: The rate of change vector is combined with the prediction time constant to obtain a prediction increment vector; The current spectral data vector is combined with the predicted increment vector to obtain the multidimensional spectral vector for the future time.

6. The method according to claim 1, characterized in that, The multidimensional spectral data vector is the spectral energy distribution vector of the surface of the bamboo and wood kitchenware in the near-infrared band.

7. The method according to claim 1, characterized in that, The predicted time constant is pre-calibrated based on the physical thermal inertia of the carbonization vessel.

8. The method according to claim 1, characterized in that, The method further includes: The final heat regulation control signal is converted into an adjustment command for the duty cycle of the pulse width modulation signal of the carbonization kettle heating unit, so as to control the energy output of the heating unit.

Images