Wood shaving mixture quality control and efficiency optimization management system

Abstract

The present application relates to the technical field of industrial process control and intelligent manufacturing, in particular to a wood shaving mixture quality control and efficiency optimization management system, comprising a performance factor deconstruction unit for collecting real-time agent attribute data to generate an agent attribute vector; converting the agent attribute vector into a standardized performance factor vector; a total factor exposure evaluation unit for calculating a real-time total factor vector; solving the factor deviation degree of the real-time total factor vector and an ideal factor vector; and discriminating the factor deviation degree to generate a qualified signal, a first-level deviation signal or a second-level deviation signal; a formula dynamic optimization unit for constructing a cost-benefit optimization function, solving a new mixing ratio and issuing an execution; a model closed-loop correction unit for constructing a performance prediction model; and updating a factor mapping matrix; the present application solves the technical problem of difficult unified management and evaluation caused by high heterogeneity of raw materials, effectively ensuring the stability of the final product quality.

Description

Technical Field

[0001] This invention relates to the field of industrial process control and intelligent manufacturing technology, specifically to a quality control and efficiency optimization management system for wood shavings mixture. Background Technology

[0002] In the field of wood shavings mixing production, the raw materials are widely available, and their physical properties and chemical characteristics exhibit high heterogeneity and uncertainty. Existing production control relies heavily on fixed mixing formulas or extensive material management, making it difficult to respond to dynamic fluctuations between batches of raw materials in a real-time and precise manner.

[0003] This approach makes it difficult to consistently target the quality of the final product, such as physical strength and stability, and also leads to difficulties in controlling production costs and material waste. Therefore, how to overcome the challenges posed by the high heterogeneity of raw materials and achieve dynamic optimization of cost-effectiveness while ensuring product quality meets standards is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides a wood shavings mixing quality control and efficiency optimization management system. Specifically, the technical solution of the present invention includes:

[0005] The performance factor deconstruction unit is used to collect real-time agent attribute data to generate agent attribute vectors; and based on the preset performance factor deconstruction model, it converts the agent attribute vectors into standardized performance factor vectors.

[0006] The total factor exposure assessment unit is used to calculate the real-time total factor vector based on the performance factor vector, the current material mixing ratio, and the ideal factor vector; calculate the factor deviation between the real-time total factor vector and the ideal factor vector; and determine the factor deviation, generating a qualified signal, a first-level deviation signal, or a second-level deviation signal.

[0007] The dynamic formulation optimization unit is used to maintain the current material mixing ratio in response to a qualified signal; and to construct a cost-benefit optimization function in response to a first-level or second-level deviation signal, solve for a new mixing ratio, and issue it for execution.

[0008] The model closed-loop correction unit is used to construct the performance prediction model; generate the final product prediction performance based on the real-time total factor vector; calculate the difference between the predicted performance and the actual quality inspection data to construct the loss function; and calculate the gradient of the loss function with respect to the factor mapping matrix in the performance factor deconstruction model to update the factor mapping matrix.

[0009] Optionally, real-time agent attribute data includes real-time moisture content, real-time geometry, and real-time procurement cost.

[0010] Optionally, the standardized performance factor vector includes: performance factor components characterizing the water absorption of the material, performance factor components characterizing the structural support, performance factor components characterizing the bonding potential, and performance factor components characterizing the cost.

[0011] Optionally, the performance factor deconstruction model includes a factor mapping matrix and a bias vector; the factor mapping matrix and the bias vector are obtained in advance through deep training combining offline physicochemical experimental data and machine learning algorithms.

[0012] Optionally, the total factor exposure assessment unit is used to: weight and sum the performance factor vector of each material with the current material mixing ratio of that material to generate a real-time total factor vector.

[0013] Optionally, the total factor exposure assessment unit is used to: calculate the vector difference between the real-time total factor vector and the ideal factor vector to generate a factor deviation vector; and calculate the magnitude of the factor deviation vector to solve for the factor deviation degree.

[0014] Optionally, the cost-benefit optimization function has an optimization objective and constraints;

[0015] The optimization objective is to minimize the total cost factor calculated by weighting the new mixing ratio with the cost factor components in the performance factor vector.

[0016] The constraints include: the deviation between the total quality factor vector calculated by the new mixing ratio and the ideal non-cost factor vector is less than the corrected allowable deviation threshold, and the sum of all new mixing ratios is 1.

[0017] Optionally, the model closed-loop correction unit is used to: adopt a performance prediction model and generate the final product prediction performance based on the real-time total factor vector; and compare the final product prediction performance with the actual quality inspection data to construct a loss function.

[0018] Optionally, the model closed-loop correction unit is also used to: apply the chain rule to backpropagate through the loss function to calculate the gradient of the factor mapping matrix; and iteratively update the factor mapping matrix according to the gradient and the preset learning rate.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. This system deconstructs real-time raw material properties from different sources and with different units into standardized performance factors, such as structural support and water absorption, through performance factor deconstruction. This solves the technical problem of difficult unified management and evaluation caused by the high heterogeneity of raw materials.

[0021] 2. This system achieves precise monitoring of production status by evaluating the deviation between the current total factor exposure of the mixture and the ideal target in real time and generating a grading signal; this provides a basis for subsequent dynamic adjustments and effectively ensures the stability of the final product quality.

[0022] 3. This system adopts a dynamic optimization unit. When a deviation is detected, it automatically constructs a cost-benefit optimization function. Under the constraint of ensuring that the total quality factor meets the standard, it can solve for the new mixing ratio with the minimum total cost factor, thus achieving the dynamic minimum cost under the premise of ensuring quality.

[0023] 4. This system constructs a model closed-loop correction unit, which uses the difference between actual quality inspection data and prediction performance to construct a loss function and backpropagates to update the deconstructed model; this enables the system to self-correct and evolve, adapt to long-term drift of raw material characteristics, and ensure the long-term robustness of the system. Attached Figure Description

[0024] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0025] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0027] Example 1:

[0028] Please see Figure 1 A wood shavings mixing quality control and efficiency optimization management system, including:

[0029] The performance factor deconstruction unit is used to collect real-time agent attribute data to generate agent attribute vectors; and based on the preset performance factor deconstruction model, it converts the agent attribute vectors into standardized performance factor vectors.

[0030] The total factor exposure assessment unit is used to calculate the real-time total factor vector based on the performance factor vector, the current material mixing ratio, and the ideal factor vector; calculate the factor deviation between the real-time total factor vector and the ideal factor vector; and determine the factor deviation, generating a qualified signal, a first-level deviation signal, or a second-level deviation signal.

[0031] The dynamic formulation optimization unit is used to maintain the current material mixing ratio in response to a qualified signal; in response to a first-level or second-level deviation signal, it constructs a cost-benefit optimization function, solves for a new mixing ratio, and issues it for execution; if the optimization solver reports no feasible solution, for example, if all available materials cannot be combined to form a qualified formulation under quality constraints, the system should immediately trigger a second-level deviation signal, maintain the current material mixing ratio unchanged or switch to a preset emergency safety formulation, and report an alarm of formulation optimization failure to management.

[0032] The model closed-loop correction unit is used to construct the performance prediction model; generate the final product prediction performance based on the real-time total factor vector; calculate the difference between the predicted performance and the actual quality inspection data to construct the loss function; and calculate the gradient of the loss function with respect to the factor mapping matrix in the performance factor deconstruction model to update the factor mapping matrix.

[0033] This embodiment provides a wood shavings mixing quality control and efficiency optimization management system. The system is a complete cyber-physical closed-loop system designed to solve the technical problems of high heterogeneity and difficulty in management of wood shavings raw materials, which leads to unstable final product quality and difficulty in cost control. The system deconstructs easily measurable raw material proxy attributes into standardized performance factors, thereby achieving real-time evaluation, dynamic optimization and model self-correction based on factor objectives.

[0034] In this embodiment, the system includes four core units:

[0035] The performance factor decomposition unit aims to convert multi-source, heterogeneous, and dimensionlessly variable real-time sensor data into unified, standardized performance indicators that can be used for management decision-making. In this embodiment, this unit collects real-time agent attribute data, such as the first, second, and third performance factors, through a near-infrared spectral (NIR) sensor, a machine vision system, and an enterprise resource planning (ERP) system. Real-time moisture content of the material Real-time geometric dimensions and real-time procurement costs This unit combines this data into a proxy attribute vector. This unit is based on a pre-defined performance factor deconstruction model, such as a Factor mapping matrix and bias vector Through mathematical operations ,Will Dimensional proxy attribute vector Convert to dimensional standardized performance factor vector ;Should The components of a vector, for example Water absorption Structural support Adhesion potential, Cost is the core basis for all subsequent decisions;

[0036] The total factor exposure assessment unit aims to monitor the overall performance status of the current blend formulation in real time and assess its deviation from the ideal target. In this embodiment, this unit obtains an ideal factor vector from the management layer. The vector The quality and cost targets for the final product are defined; simultaneously, this unit is based on each... arrive Material performance factor vector and the current material mixing ratio obtained from the production line PLC / DCS system Through weighted summation To calculate the real-time total factor vector ; This unit characterizes the total factor exposure of the current mixture; it calculates... and Vector difference between And further calculate the magnitude of the deviation vector, such as the L2 norm. This yields a scalar factor deviation. The deviation of the discriminant factor in this unit :like , If the threshold is met, a pass signal is generated; if... , If the first-level deviation threshold is met, then a first-level deviation signal is generated; if Then a second-level deviation signal is generated; where the threshold is... and The value can be based on historical production data. Statistical analysis of the distribution, such as taking the 95th percentile or setting it in conjunction with product quality tolerance standards;

[0037] The purpose of the dynamic formulation optimization unit is to automatically find a new optimal material mixing ratio when the system deviates from the ideal target, so as to achieve a balance between cost and benefit. In this embodiment, the unit operates in response to received signals: if a qualified signal is received, the current material mixing ratio is maintained. The process remains unchanged; if a first-level or second-level deviation signal is received, the optimization engine is immediately triggered; this engine constructs a cost-benefit optimization function that minimizes the total cost factor. To optimize the objective while ensuring that the deviation of the total quality factor is within the allowable range, for example... And the sum of the proportions is 1 As constraints, this unit solves the function using standard optimization algorithms, such as linear programming or quadratic programming, to obtain a new set of mixing ratios. It then sends the instructions to the underlying PLC / DCS system for execution, adjusting the feeding ratio.

[0038] The purpose of the model closed-loop correction unit is to address the performance factor deconstruction model caused by long-term drift in raw material properties, such as seasonal changes. To address the issue of failures and ensure the system's long-term robustness, in this embodiment, the unit constructs a performance prediction model. The model Able to base on real-time total factor vector To generate predicted performance of the final product ,Right now After the product manufacturing is completed and actual quality inspection data is obtained. For example, the actual physical strength and density, this unit calculates and predicts performance. and The difference is used to construct a loss function. This unit applies backpropagation and the chain rule to calculate the loss function. Factor mapping matrix in performance factor deconstruction model gradient Using gradient descent, the gradient and the preset learning rate are utilized. To update the factor mapping matrix: ;

[0039] The wood shavings mixing quality control and efficiency optimization management system described in this embodiment constructs a complete technical closed loop through the collaborative work of the four main units. This loop extends from the real-time raw material perception and deconstruction unit to the production status assessment unit, then to the dynamic optimization and seeking unit for decision-making, and finally back to the model self-evolution and correction unit. It solves the problems of quality control difficulties and cost waste caused by the high heterogeneity of raw materials in existing technologies. It achieves the goal of dynamically minimizing cost factors while ensuring that the final product quality factors, such as strength and stability, are always anchored to the ideal target, thereby achieving the dual goals of quality control and efficiency optimization.

[0040] Example 2:

[0041] Real-time agent attribute data includes real-time moisture content, real-time geometric dimensions, and real-time procurement costs.

[0042] This embodiment is a specific limitation on the real-time agent attribute data collected by the performance factor deconstruction unit in Embodiment 1;

[0043] Real-time surrogate attribute data refers to raw data that is easily measurable in real time and is used to indirectly characterize the final properties of materials; in this embodiment, they constitute the surrogate attribute vector. Main components:

[0044] Real-time moisture content :

[0045] This refers to the weight percentage of moisture contained within the wood shavings material;

[0046] Moisture content is one of the most important parameters in wood processing, as it directly affects subsequent gluing effects, drying energy consumption, and the dimensional stability of the final product.

[0047] In this embodiment, the data is acquired in real time and continuously in a non-contact manner by a near-infrared spectroscopy (NIR) sensor installed above the conveyor belt.

[0048] Real-time geometry :

[0049] It refers to the morphological characteristics of wood shavings, which can be a vector including statistical values ​​such as average length, width, thickness, or length-to-thickness ratio;

[0050] The geometric dimensions determine the layout of the wood shavings and the contact area between the particles during hot pressing, which directly affects the physical and mechanical strength of the final product, such as static bending strength and internal bond strength.

[0051] In this embodiment, the data is obtained in real time by the machine vision system through high-speed photography and image processing algorithms;

[0052] Real-time procurement costs :

[0053] This refers to the purchase price or total cost per unit weight or volume of the batch of materials when it enters the production line.

[0054] This data is a key economic input for the subsequent dynamic optimization unit of the formulation to perform cost-benefit optimization;

[0055] In this embodiment, the data can be automatically obtained from the material management module interface of the enterprise ERP system, or manually entered by the operator based on the purchase order;

[0056] By using these three specific real-time data points—moisture content, geometric dimensions, and procurement cost—as core inputs for proxy attributes, this invention ensures that the performance factor deconstruction unit can model from two dimensions: the key physical characteristic affecting the final product performance—moisture content—geometric dimensions, and the key economic characteristic—procurement cost. Compared to using only single or incomplete attribute data, the inputs in this embodiment provide the most direct and sufficient original basis for subsequently deconstructing accurate performance factors such as water absorption, structural support, and cost factors, greatly improving the accuracy of model deconstruction and the feasibility and effectiveness of subsequent cost optimization.

[0057] Example 3:

[0058] The standardized performance factor vector includes: performance factor components characterizing material water absorption, performance factor components characterizing structural support, performance factor components characterizing bonding potential, and performance factor components characterizing cost.

[0059] This embodiment is a specific definition of the standardized performance factor vector output by the performance factor deconstruction unit in Embodiment 1;

[0060] Standardized performance factor vector This refers to heterogeneous proxy attributes By deconstructing the model The transformed decision-making basis vector has unified dimensions and clear physical or economic significance; in this embodiment, It is dimensional vectors, for example Its weight Specifically, it includes:

[0061] Performance factor components characterizing the water absorption of materials :

[0062] This is a standardized scalar value that characterizes the material's tendency to expand due to moisture absorption under specific conditions such as humidity and temperature.

[0063] This factor is used to predict and control the water absorption swelling rate and dimensional stability of the final product;

[0064] It is mainly determined by the real-time moisture content in the proxy attribute. and real-time geometry For example, specific surface area, through a mapping matrix The result is obtained through comprehensive deconstruction;

[0065] Performance factor components characterizing structural support :

[0066] This is a standardized scalar value that characterizes the physical and mechanical support capabilities that the material can provide after being pressed into a sheet.

[0067] This factor is used to predict and control the core mechanical properties of the final product, such as static bending strength and elastic modulus.

[0068] It is mainly composed of real-time geometric dimensions. For example, the length-to-thickness ratio, through the mapping matrix Deconstruction yields;

[0069] Performance factor components characterizing bonding potential :

[0070] This is a standardized scalar value that characterizes the ease and strength of chemical or physical bonding between the surface properties of the material and the adhesive.

[0071] This factor is used to predict and control the internal bonding strength of the final product and the possible formaldehyde release.

[0072] It can be determined by real-time moisture content. Both excessively high and low values ​​can affect bonding, as well as other possible proxy properties such as the wood chemical composition reflected in the NIR spectrum via matrix. Deconstruction yields;

[0073] Performance factor components characterizing cost :

[0074] This is a standardized scalar value that represents the overall economic cost of using the material.

[0075] This factor is the main optimization objective of the cost-benefit optimization function in the dynamic optimization unit of the formulation;

[0076] It is mainly composed of real-time procurement costs. Through matrix Deconstruction reveals that, in a simplified embodiment, Possibly with The relationship is linear or equal, but the deconstruction model provides the possibility of handling more complex cost structures, such as considering drying energy consumption;

[0077] By converting heterogeneous and difficult-to-manage proxy attributes such as moisture content (%), dimensions (mm), and cost (yuan / ton) into four standardized performance factors directly linked to the final product's water absorption, support, bonding, and cost-effectiveness, this invention solves the core technical challenge of evaluating and optimizing different types of data within the same framework. It enables managers to set quality objectives for the first time on a unified factor dimension, such as... Production control is a necessary prerequisite for subsequent total factor exposure assessment and dynamic formulation optimization.

[0078] Example 4:

[0079] The performance factor deconstruction model includes a factor mapping matrix and a bias vector; the factor mapping matrix and the bias vector are obtained in advance through deep training combining offline physicochemical experimental data and machine learning algorithms.

[0080] This embodiment is a detailed description of the mathematical structure of the preset performance factor decomposition model used by the performance factor decomposition unit in Embodiment 1 and the source of its parameters;

[0081] Mathematical structure of the model:

[0082] In order to achieve from Dimensional proxy attribute vector arrive dimensional performance factor vector For efficient mapping, the performance factor deconstruction model in this embodiment adopts a linear affine transformation structure, the mathematical expression of which is as follows:

[0083] ;

[0084] For the first The performance factor vector of a material The dimensional vector is calculated from this model;

[0085] For the first The proxy attribute vector of a material. A dimensional vector, acquired by the performance factor deconstruction unit, such as... ;

[0086] The factor mapping matrix, The dimensional matrix, a core parameter of this model, is obtained through pre-training.

[0087] For bias vectors, The dimensional vector, the core parameter of this model, is obtained through pre-training.

[0088] Technical Motivation: This linear model structure achieves high computational efficiency, requiring only one matrix multiplication and addition to meet the real-time deconstruction needs in the production process; furthermore, the matrix... Each element in All of these intuitively reflect the first Each proxy attribute, such as For the Performance factors such as The contribution weights are somewhat interpretable;

[0089] Source of model parameters:

[0090] One of the core aspects of this invention lies in the factor mapping matrix. and bias vector It is not calculated in real time or set manually, but rather obtained in advance; the acquisition process is as follows:

[0091] Collect a large number of samples, for example, 1000 sets of wood shavings; for each sample On the one hand, measure its proxy attribute vector For example, using NIR and machine vision to measure its On the other hand, its true performance factor is measured through offline physicochemical experiments. For example, the water absorption swelling rate test was conducted according to national standards. Mechanical tests were conducted to obtain etc.; this This will be used as a label for training;

[0092] Collected The dataset is used as the training set;

[0093] Apply machine learning algorithms, such as least squares for multiple linear regression, or use shallow neural networks to fit the data. Relationships, solve for the optimal matrix sum vector , making Minimum;

[0094] This embodiment illustrates a scientific and robust model construction method; by combining offline physicochemical experimental data, the model's reliability is ensured. and The represented mapping relationship has a solid physicochemical basis, rather than being a pure data fit; by applying machine learning algorithms for deep training, the model can automatically learn and solidify the complex high-dimensional linear coupling relationship between agent attributes and performance factors from a large amount of data; this mode of offline deep training and online efficient application not only ensures the accuracy of model deconstruction, but also meets the efficiency requirements of real-time control, which is the cornerstone for the reliable operation of the entire system.

[0095] Example 5:

[0096] The total factor exposure assessment unit is used to: weight and sum the performance factor vector of each material with the current material mixing ratio to generate a real-time total factor vector.

[0097] This embodiment defines the specific mathematical method by which the total factor exposure assessment unit in Embodiment 1 calculates the real-time total factor vector;

[0098] Real-time total factor vector Its purpose is to quantify the results of... The overall performance of the final mixture after mixing different materials according to the current production formula; in this embodiment, the calculation logic draws on the principle of calculating total risk exposure in financial asset portfolio management, that is, the total factor exposure of the mixture is the weighted average of the factor exposures of each individual material that constitutes it, and the calculation method is as follows:

[0099] ;

[0100] This represents the real-time total factor vector of the final mixture. A dimensional vector, obtained by this calculation; each of its components is as follows: This represents the comprehensive exposure value of the current mixture in terms of four factors: water absorption, support, bonding potential, and cost.

[0101] For the first The performance factor vector of a material The dimension vector is provided in real time by the performance factor decomposition unit;

[0102] For the first The current mixing ratio of the materials, scalar quantity, and This data is obtained in real time by the total factor exposure assessment unit from the underlying PLC / DCS execution layer;

[0103] This represents the total number of material types.

[0104] The core logic of this calculation lies in the assumption that each performance factor of the final mixture, such as total water absorption, is assumed to be... It is the corresponding performance factor of each material that makes it up. According to their respective mixing ratios The result of weighted summation is This assumption of linear superposition is a reasonable and efficient approximation in many mixed material applications.

[0105] By employing this weighted summation method, this invention provides a scientific, quantitative, and computationally efficient method for real-time evaluation of the total factor exposure of a material combination. It enables managers to monitor production status in real time for the first time from a unified factor dimension rather than the traditional, difficult-to-manage ratio dimension. As a The vector accurately marks the current production state point. The position in the performance space is used to subsequently compare it with the ideal target. The comparison and evaluation of deviations provide a precise basis for calculation.

[0106] Example 6:

[0107] The total factor exposure assessment unit is used to: calculate the vector difference between the real-time total factor vector and the ideal factor vector to generate the factor deviation vector; and calculate the magnitude of the factor deviation vector to solve for the factor deviation degree.

[0108] This embodiment defines the specific calculation steps for how the total factor exposure assessment unit in Embodiment 1 quantifies the key indicator of factor deviation.

[0109] The real-time total factor vector was calculated using the method described in Example 5. Then, the system needs to compare it with the ideal factor vector preset by the management. Compare to generate factor deviation vectors ;

[0110] Ideal factor vector : The dimension vector originates from the management's pre-set benchmark target for production based on the final product standards.

[0111] ;

[0112] This is the factor deviation vector. The dimension vector is obtained from this calculation;

[0113] The real-time total factor vector is calculated using the method described in Example 5;

[0114] Ideal factor vector, pre-defined;

[0115] The technical significance lies in that it On each performance factor dimension, specific deviations between the current production status and the ideal target are defined; for example... The first component If the value is positive, it indicates that the total water absorption factor of the current mixture exceeds the standard; if it is negative, it indicates that the standard is not met. This vector... Information regarding the direction and magnitude of the deviation was preserved;

[0116] To obtain an overall scalar of deviation for graded assessment, this embodiment calculates the factor deviation vector. The modulus, such as the L2 norm (Euclidean distance), is used to calculate the factor deviation. ;

[0117] ;

[0118] Factor deviation, a scalar, is calculated in this way;

[0119] Factor deviation vector The One component;

[0120] The technical meaning is Vector and Vector in Geometric distance in the dimensional performance factor space; The larger the value, the more severe the overall deviation of the current production state from the ideal target; this scalar... Simple and intuitive, it is very suitable for subsequent threshold determination, for example. and and Comparison;

[0121] This embodiment calculates the vector difference first. Then calculate the modulus. This method enables precise quantification of deviations; Vectors retain detailed deviation information across all dimensions, which can be used for more refined control logic in the future, such as adjusting only specific factors with severe deviations; while The scalar provides a clear overall deviation that can be used to classify and distinguish between qualified, first-level, and second-level standards; this design balances the completeness of information with the convenience of decision-making, and provides a reliable input for subsequent evaluation and optimization units.

[0122] Example 7:

[0123] The cost-benefit optimization function has an optimization objective and constraints;

[0124] The optimization objective is to minimize the total cost factor calculated by weighting the new mixing ratio with the cost factor components in the performance factor vector.

[0125] The constraints include: the deviation between the total quality factor vector calculated by the new mixing ratio and the ideal non-cost factor vector is less than the corrected allowable deviation threshold, and the sum of all new mixing ratios is 1.

[0126] This embodiment provides a detailed definition of the core mathematical structure of the cost-benefit optimization function constructed by the dynamic optimization unit for the formula in Embodiment 1; this function embodies the management philosophy of minimizing costs while ensuring quality;

[0127] When the total factor exposure assessment unit triggers a first- or second-level deviation signal, the system automatically activates the formulation hedging engine; the core of this engine is to solve a constrained optimization problem to find a new set of mixing ratios in real time. The cost-benefit optimization function is constructed as follows:

[0128] Optimization target: ObjectiveFunction

[0129] The optimization objective is to minimize the total cost factor calculated by weighting the new mixing ratio with the cost factor components in the performance factor vector.

[0130] ;

[0131] The first one to be solved A new mixing ratio for a type of material, scalar;

[0132] For the first The cost factor components of a material, scalars, originate from The Each component is provided by the performance factor decomposition unit;

[0133] The optimization objective This is the total cost factor under the new formula, and the goal is to minimize it to achieve cost arbitrage.

[0134] Constraints:

[0135] The constraints include: and the sum of all new mixing ratios is 1;

[0136] (1) as well as ;

[0137] This constraint, where the sum of the proportions is 1, ensures the physical meaning of the ratio.

[0138] The constraints include: the deviation between the total quality factor vector calculated by the new mixing ratio and the ideal non-cost factor vector is less than the corrected allowable deviation threshold;

[0139] (2) ;

[0140] For the first The non-cost performance factor vector of the material, Dimensional vector; Term definition: refers to Cost factors have been removed. The remaining vectors, for example This vector is collectively referred to as the quality factor vector;

[0141] For an ideal non-cost factor vector, Dimensional vector; Term definition: refers to The vector remaining after removing the cost factor represents the quality target of the final product.

[0142] The revised allowable deviation threshold is a scalar; terminology definition: its source is an adjustable parameter set by management based on quality tolerance, and its determination method can be adjusted based on the stringency of the required product grade, for example, higher grade products correspond to smaller tolerances. This defines the maximum permissible deviation of the new formulation from the total quality factor;

[0143] This is the core quality constraint; This is the total quality factor vector under the new formulation; the meaning of this constraint is: the total quality factor vector of the new formulation is close to the ideal quality target. The deviation between them is a scalar, and must be strictly controlled within the new threshold. Within;

[0144] The optimization function defined in this embodiment mathematically precisely achieves the goal of finding the material combination with the lowest cost optimization objective while ensuring that the total quality factor constraint 2 is met; it completely abandons the rigid thinking of fixed formulas and turns to the management philosophy of fixed target factor quality; when raw materials Fluctuations occur regardless of performance Volatility or Cost When fluctuations cause the old formula to deviate, the optimization unit can automatically and in real time calculate a new optimal formula. This allows production to be brought back to the quality standard range at minimal cost, thereby greatly improving economic efficiency and system adaptability while ensuring quality.

[0145] Example 8:

[0146] The model closed-loop correction unit is used to: adopt a performance prediction model and generate the final product prediction performance based on the real-time total factor vector; and compare the final product prediction performance with the actual quality inspection data to construct a loss function.

[0147] This embodiment is a detailed explanation of how the model closed-loop correction unit in Embodiment 1 establishes a connection between the model and reality and quantifies their differences; this is a prerequisite step for realizing closed-loop correction.

[0148] The model closed-loop correction unit is used to: adopt a performance prediction model and generate the final product prediction performance based on the real-time total factor vector;

[0149] To address the performance factor deconstruction model caused by long-term drift in raw material properties To address the issue of failure, this embodiment introduces a performance prediction model. ;

[0150] Performance prediction model :

[0151] This refers to an independent mathematical model, such as a multiple regression model, a neural network, or a support vector machine.

[0152] Its function is to establish the real-time total factor vector. Internal management metrics of the system and the predicted performance of the final product The final physical results and the mathematical relationships between them;

[0153] The model Also obtained through offline training, for example, using a large amount of historical production data. The data was obtained by fitting the data pairs.

[0154] ;

[0155] For predicting the performance of the final product, vectors such as predicting intensity and predicting density are used. By this model Calculated;

[0156] Real-time total factor vector The dimension vector is calculated using the method described in Example 5;

[0157] This step uses a model. The system will be evaluated in factor space. Predictions mapped to the actual performance space The system then compares the predicted performance of the final product with the actual quality inspection data to construct a loss function; after the corresponding batch of products is manufactured and enters the quality inspection (QC) stage, the system obtains its actual quality inspection data. and compared it with the system's earlier predictions. A comparison is performed to quantify the accuracy of the predictions; this quantified difference is the loss function. ;

[0158] Actual quality inspection data :

[0159] This refers to authoritative performance data obtained by a QC laboratory through physical or chemical testing;

[0160] For example, actual strength and actual density ;

[0161] ;

[0162] ;

[0163] Loss function, a scalar, is calculated here;

[0164] The form of the loss function, and the basis for its value: can be selected according to the specific task, such as mean squared error (MSE). ;

[0165] This embodiment introduces a performance prediction model. In performance factor Internal management indicators and actual product performance A crucial bridge has been built between the final quality results; without It is impossible to deviation and The deviations are correlated; by constructing a loss function The system can be represented by a definite scalar. To quantify its internal model, including and The accuracy of predictions about the real world; The larger the value, the further the model deviates from reality; this The value is the error signal source that initiates subsequent backpropagation and model correction, and is the basis for achieving closed-loop adaptation.

[0166] Example 9:

[0167] The model closed-loop correction unit is also used to: apply the chain rule to calculate the gradient of the factor mapping matrix through backpropagation of the loss function; and iteratively update the factor mapping matrix according to the gradient and the preset learning rate.

[0168] This embodiment, based on Embodiment 8, focuses on how the model closed-loop correction unit utilizes the pre-constructed loss function. To specifically correct the most fundamental factor mapping matrix of the system. The detailed explanation constitutes the core execution steps of the closed-loop correction;

[0169] The model closed-loop correction unit is also used to: apply the chain rule to calculate the gradient of the factor mapping matrix through backpropagation of the loss function;

[0170] In order to Error backpropagation to correct This embodiment uses the gradient descent algorithm; loss needs to be calculated. For matrix The partial derivative, i.e., the gradient, ;

[0171] This gradient calculation must apply the chain rule because and There exists a long dependency chain:

[0172] Depends on See Example 8. ;

[0173] Depends on See Example 8. ;

[0174] Depends on See Example 5. ;

[0175] Depends on See Example 4. According to the matrix differentiation rule;

[0176] By using the chain rule, multiply the above gradient terms together. The system can calculate the final gradient matrix. The physical meaning of this gradient matrix is: Each element in In which direction, positive or negative, should the adjustment be made to minimize the loss? It dropped the fastest;

[0177] The factor mapping matrix is ​​iteratively updated based on the gradient and the preset learning rate.

[0178] exist Calculate the gradient at each time step Then, the system uses the gradient descent method to... Perform iterative updates to generate The new matrix of time :

[0179] ;

[0180] The updated factor mapping matrix. 3D matrix;

[0181] The factor mapping matrix at the current time. 3D matrix;

[0182] Learning rate, a scalar; Term definition: This is a pre-defined hyperparameter, for example... The value is determined by debugging or cross-validation. Its function is to control the step size of each update to prevent oscillation caused by updating too quickly or slow convergence caused by updating too slowly.

[0183] Loss function on matrix gradient, The dimensional matrix is ​​calculated using the chain rule;

[0184] This embodiment, together with Embodiment 8, constitutes a complete system based on actual quality data. The model closed-loop correction system utilizes the chain rule and gradient descent, taking advantage of the final quality check results. To reverse-engineer the core parameters in the original performance factor deconstruction model. The remarkable advantage of this design lies in its ability to automatically adapt the system to long-term drift of raw materials, such as due to seasonal changes. Corresponding to different ;when The failure began When the value increases, the closed-loop correction unit automatically adjusts it back to the correct relationship; this ensures the long-term effectiveness, accuracy and robustness of the entire management system and avoids the common problem of the model failing due to environmental changes after deployment.

[0185] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A quality control and efficiency optimization management system for wood shavings mixing, characterized in that, include: The performance factor deconstruction unit is used to collect real-time agent attribute data to generate agent attribute vectors; Based on a pre-defined performance factor deconstruction model, the proxy attribute vector is converted into a standardized performance factor vector. The total factor exposure assessment unit is used to calculate the real-time total factor vector based on the performance factor vector, the current material mixing ratio, and the ideal factor vector; Calculate the factor deviation between the real-time total factor vector and the ideal factor vector; It also determines the degree of factor deviation and generates a qualified signal, a first-level deviation signal, or a second-level deviation signal; The dynamic formulation optimization unit is used to maintain the current material mixing ratio in response to a qualified signal; and to construct a cost-benefit optimization function in response to a first-level or second-level deviation signal, solve for a new mixing ratio, and issue it for execution. The model closed-loop correction unit is used to construct the performance prediction model; generate the final product prediction performance based on the real-time total factor vector; and calculate the difference between the predicted performance and the actual quality inspection data to construct the loss function. The gradient of the loss function with respect to the factor mapping matrix in the performance factor deconstruction model is calculated to update the factor mapping matrix.

2. The wood shavings mixing quality control and efficiency optimization management system according to claim 1, characterized in that, The real-time agent attribute data includes real-time moisture content, real-time geometric dimensions, and real-time procurement cost.

3. The wood shavings mixing quality control and efficiency optimization management system according to claim 1, characterized in that, The standardized performance factor vector includes: performance factor components characterizing material water absorption, performance factor components characterizing structural support, performance factor components characterizing bonding potential, and performance factor components characterizing cost.

4. The wood shavings mixing quality control and efficiency optimization management system according to claim 1, characterized in that, The performance factor deconstruction model includes a factor mapping matrix and a bias vector; the factor mapping matrix and the bias vector are obtained in advance through deep training by combining offline physicochemical experimental data with machine learning algorithms.

5. The wood shavings mixing quality control and efficiency optimization management system according to claim 1, characterized in that, The total factor exposure assessment unit is used to: weight and sum the performance factor vector of each material with the current material mixing ratio to generate a real-time total factor vector.

6. The wood shavings mixing quality control and efficiency optimization management system according to claim 1, characterized in that, The total factor exposure assessment unit is used to: calculate the vector difference between the real-time total factor vector and the ideal factor vector to generate a factor deviation vector; and calculate the magnitude of the factor deviation vector to solve for the factor deviation degree.

7. The wood shavings mixing quality control and efficiency optimization management system according to claim 1, characterized in that, The cost-benefit optimization function has an optimization objective and constraints; The optimization objective is to minimize the total cost factor calculated by weighting the new mixing ratio with the cost factor components in the performance factor vector. The constraints include: the deviation between the total quality factor vector calculated by the new mixing ratio and the ideal non-cost factor vector is less than the corrected allowable deviation threshold, and the sum of all new mixing ratios is 1.

8. The wood shavings mixing quality control and efficiency optimization management system according to claim 1, characterized in that, The model closed-loop correction unit is used to: adopt a performance prediction model and generate the final product prediction performance based on the real-time total factor vector; and compare the final product prediction performance with the actual quality inspection data to construct a loss function.

9. The wood shavings mixing quality control and efficiency optimization management system according to claim 8, characterized in that, The model closed-loop correction unit is also used to: apply the chain rule to calculate the gradient of the factor mapping matrix through backpropagation of the loss function; and iteratively update the factor mapping matrix according to the gradient and the preset learning rate.

Images