Split toilet production process control method and system, and storage medium

Abstract

The application provides a split toilet production process control method and system and a storage medium. Real-time state parameters of each process of a split toilet to be processed are collected, historical state parameter sets of previous qualified products and recently processed products are combined, and a space-time state matrix is constructed. The matrix is subjected to feature decomposition and dimension reduction to generate a feature subspace covering main process fluctuation information. The composite fault contribution weight of each parameter is calculated according to the parameter batch statistical dispersion and the current batch energy spectrum density integral value. The weight and the preset fault feature basis vector representing the crack, deformation and pinhole defects are subjected to tensor contraction and operation to obtain a predicted fault vector. The initial process control instruction is obtained by solving the minimum weighted sum of the norm of the predicted fault vector and the norm of the process control parameter adjustment amount. If any parameter adjustment amount in the initial instruction exceeds the threshold value, an asymmetric penalty term related to the adjustment direction is applied, and the process control instruction is obtained by re-solving.

Description

Technical Field

[0001] This application belongs to the field of control, and in particular relates to a method, system and storage medium for controlling the manufacturing process of a split toilet. Background Technology

[0002] The production process control of split-type toilets relies on operator experience and offline quality inspection, which is highly subjective, slow to respond, and difficult to cope with real-time production environments. Statistical process control methods monitor key parameters, but these are often univariate or low-dimensional, making it difficult to detect the correlation between multiple processes and the synergistic effects of numerous parameters. They are also insensitive to slow drift and potential fault signs caused by multiple variables, failing to achieve early prediction and proactive intervention in product quality. Data-driven multivariate statistical analysis methods, such as principal component analysis and partial least squares, monitor the overall process status by reducing the dimensionality of high-dimensional process data. However, when building models, they fail to fully integrate historical qualified product data and batch-to-batch fluctuation information, resulting in insufficient representation of long-term process stability and batch-to-batch variability. At the fault diagnosis level, while existing methods can detect process anomalies, their diagnostic logic is relatively simplistic when representing the contribution of each state parameter to specific defect types, failing to distinguish the influencing factors leading to different types of defects. When generating control strategies, feedback control models lack comprehensive prediction and trade-offs for multiple potential failure risks. They do not consider that control adjustments should have different costs and penalty mechanisms under different risk levels, and therefore cannot achieve optimal control under the balance of risk and cost. Summary of the Invention

[0003] This invention proposes a manufacturing process control method for split-type toilets to address the problems of existing methods failing to fully integrate historical qualified product data and batch-to-batch fluctuation information, failing to distinguish the influencing factors causing different types of defects, and failing to achieve optimal control under a balance of risk and cost. The method includes: The system acquires the real-time status parameters of the split toilet to be processed at each stage, and combines them with the historical status parameter sets of multiple qualified products from the previous stage and recently processed toilets to construct a spatiotemporal state matrix. The spatiotemporal state matrix is ​​subjected to eigenvalue decomposition and dimensionality reduction to construct a feature subspace containing information on the main process fluctuations. Within the feature subspace, the composite fault contribution weight of each state parameter is calculated, wherein the weight is positively correlated with the statistical dispersion of the state parameter across batches and the energy spectral density integral value within the current batch. The composite fault contribution weight is combined with the preset fault feature basis vectors representing cracks, deformations and pinholes respectively by tensor shrinkage operation to obtain a predicted fault vector containing multiple potential fault probabilities; the initial process control command is obtained by solving an optimization model with the objective function of minimizing the norm of the predicted fault vector and the norm of the process control parameter adjustment. When the adjustment amount of any parameter in the initial process control command exceeds the threshold, an asymmetric penalty term related to the adjustment direction is applied to the command and the solution is re-solved to obtain the process control command.

[0004] Furthermore, the present invention also relates to a manufacturing process control system for split toilets, comprising the following modules: The module is used to obtain the real-time status parameters of the current split toilet to be processed in each process, and to construct a spatiotemporal state matrix by combining the historical status parameter sets of multiple previous qualified products and recently processed toilets. The calculation module is used to perform eigenvalue decomposition and dimensionality reduction on the spatiotemporal state matrix to construct a feature subspace containing the main process fluctuation information; within the feature subspace, the composite fault contribution weight of each state parameter is calculated, wherein the weight is positively correlated with the statistical dispersion of the state parameter across batches and the energy spectral density integral value in the current batch; The first solution module is used to perform tensor shrinking operation on the composite fault contribution weight and the preset fault feature basis vectors representing cracks, deformations and pinhole defects respectively, to obtain a predicted fault vector containing multiple potential fault probabilities; and to obtain the initial process control command based on an optimization model with the objective function of minimizing the norm of the predicted fault vector and the norm of the process control parameter adjustment amount. The second solution module is used to apply an asymmetric penalty term related to the adjustment direction to the instruction and re-solve when the adjustment amount of any parameter in the initial process control instruction exceeds the threshold, so as to obtain the process control instruction.

[0005] This invention constructs a spatiotemporal state matrix that integrates current batch, historical qualified product, and recent production data, enabling it to represent a panoramic view of the production process and detect long-term fluctuations and subtle differences between batches. The method for calculating composite fault contribution weights combines cross-batch dispersion of parameters with intra-batch energy spectral density, achieving traceability of fault sources. Using tensor contraction operations, it can predict the probability of various specific defects in advance, transforming quality control from post-event detection to pre-event prevention. When formulating control strategies, the optimized model balances the goals of suppressing fault risks and reducing control adjustment amplitude. Utilizing an asymmetric penalty mechanism, it can modify control commands based on predicted main defect types and related processes, improving the accuracy of quality defect control and increasing the yield rate of split-type toilet production. Attached Figure Description

[0006] Figure 1 A flowchart of the first embodiment; Figure 2 This is a schematic diagram of a two-piece toilet. Figure 3 A schematic diagram illustrating the contribution weights of composite faults to various state parameters; Figure 4 This is a schematic diagram illustrating the relationship between threshold and fault risk. Figure 5 This is a diagram comparing the initial and final instructions. Detailed Implementation

[0007] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0008] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0009] Figure 1 This is a flowchart of the first embodiment of the present invention, namely a method for controlling the manufacturing process of a split toilet, which specifically includes: S1: Obtain the real-time status parameters of the split toilet to be processed in each process, and combine them with the historical status parameter sets of multiple previous qualified products and recently processed toilets to construct a spatiotemporal state matrix. Figure 2This is a schematic diagram of a two-piece toilet, including a water tank assembly, seat ring fastener, flush valve, and drain pipe. In one embodiment, the manufacturing process of the two-piece toilet includes, but is not limited to, the manufacturing of the toilet body and the manufacturing of various accessories. Accessories include the aforementioned water tank assembly. Taking the manufacturing of the toilet body as an example, sensors such as densitometers, temperature sensors, pressure sensors, hygrometers, and viscometers are installed during the grouting, drying, glazing, and firing processes to collect data in real time. The state parameters include, for example, the slurry density and temperature during the grouting process, multiple sampling points on the temperature and humidity curve during the drying process, the glaze viscosity during the glazing process, and the heating rate and holding time during the firing process. Simultaneously, the same state parameters of the 10 most recently produced batches of qualified products and all toilets from the 20 most recently produced batches are retrieved from a historical database. The real-time parameter data matrix of the current batch up to the current moment, the data matrix of historical qualified products, and the data matrix of recently processed toilets are spliced ​​along the batch dimension to obtain a two-dimensional spatiotemporal state matrix where rows represent samples from different batches and time points, and columns represent different state parameters. The following example focuses on the processing of the toilet body. In one embodiment, the present invention is particularly applicable to the production process control of the split toilet body, i.e., the toilet base.

[0010] To construct a data structure that reflects the temporal and spatial characteristics of the process state, in an optional embodiment, the real-time state parameters of the currently processed split toilet at each stage are obtained, and combined with the historical state parameter sets of multiple preceding qualified products and recently processed toilets, a spatiotemporal state matrix is ​​constructed, including: The four real-time status parameters of the split toilet to be processed are: grouting pressure, slurry temperature, slurry intake time, and ambient humidity. These parameters are used as one row of a matrix. Select the four status parameters of the first 10 qualified products in the corresponding process as 10 rows; We selected 50 recently processed toilets and used 4 state parameters of the corresponding process as 50 rows to form a 61×4-dimensional spatiotemporal state matrix.

[0011] The matrix integrates information from three key components: the instantaneous state of the current product, the baseline of stable production, and the recent fluctuation history of the process. The model structure is a 61-row, 4-column matrix, where each row represents a production sample and each column represents a process parameter. These four parameters are grouting pressure, slurry temperature, slurry intake time, and ambient humidity. Alternatively, they could be drying chamber temperature, drying chamber humidity, firing kiln temperature, and firing time.

[0012] Four real-time parameters of the currently processed split-type toilets are collected, such as grouting pressure of 1.2 MPa, grouting time of 900 seconds, slurry temperature of 32℃, and ambient humidity of 50%. This set of data constitutes the first row of the spatiotemporal state matrix, representing the current state at the point in time that needs to be evaluated and controlled. The corresponding process parameters of the 10 most recently produced qualified products without any defects are retrieved from the historical database. These 10 sets of data constitute rows 2 to 11 of the matrix, collectively defining an ideal and stable process state window. Then, the process parameters of the 50 most recently processed toilets, regardless of their quality, are used as rows 12 to 61 of the matrix. This part of the data contains information on possible drift, fluctuations, and disturbances that may occur during the process. These three parts of data are then vertically concatenated to form a 61×4-dimensional comprehensive state matrix. Those skilled in the art should understand that the number of state parameters is not limited to four; more can be used, and it is not limited to the four state parameters of the 50 most recently processed toilets and the first 10 qualified products at the corresponding process stage.

[0013] S2, Perform eigenvalue decomposition and dimensionality reduction on the spatiotemporal state matrix to construct a feature subspace containing the main process fluctuation information; Within the feature subspace, calculate the composite fault contribution weight of each state parameter, wherein the weight is positively correlated with the statistical dispersion of the state parameter across batches and the energy spectral density integral value in the current batch; The spatiotemporal state matrix, i.e., each state parameter, is standardized. Principal component analysis (PCA) is used to calculate the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvectors corresponding to eigenvalues ​​with a cumulative contribution rate exceeding 95% are selected to obtain the eigenspace. For each state parameter, the variance of that parameter across all batches of data is calculated as the cross-batch statistical dispersion. Then, a Fast Fourier Transform (FFT) is performed on the time series data of that parameter within the current batch to calculate the energy spectral density function. This function is then integrated over the entire frequency domain to obtain the energy spectral density integral value. The cross-batch statistical dispersion of that parameter is weighted and summed with the energy spectral density integral value to obtain the composite fault contribution weight of that parameter. The weighting coefficients are determined based on expert experience or regression analysis of historical data.

[0014] To extract low-dimensional features that represent the main fluctuations of the system from high-dimensional raw data containing noise and redundant information, in an optional embodiment, the step of performing eigenvalue decomposition to reduce the dimensionality of the spatiotemporal state matrix and constructing a feature subspace containing information on the main process fluctuations includes: The spatiotemporal state matrix is ​​processed using principal component analysis. Calculate the covariance matrix of the matrix, and solve for the eigenvalues ​​and eigenvectors of the covariance matrix; The eigenvectors corresponding to the k largest eigenvalues ​​with a cumulative variance contribution rate of 95% are selected to form the feature subspace.

[0015] Specifically, the model used is the classic principal component analysis model, which transforms the original four related process parameters into a few linearly independent principal components through linear transformation. These principal components can retain the variation information of the original data.

[0016] The constructed 61×4 dimensional spatiotemporal state matrix is ​​centered by subtracting the mean from each column. A 4×4 dimensional covariance matrix is ​​then calculated, where the diagonal elements represent the variance of each parameter individually, and the off-diagonal elements represent the covariance between any two parameters. Eigenvalue decomposition is performed on this covariance matrix to obtain four eigenvalues ​​and their corresponding four eigenvectors. For example, the calculated eigenvalues ​​are sorted by size as follows: , , , The cumulative variance contribution rate was calculated. The contribution rate of the first principal component was approximately 64%, and the cumulative contribution rate of the first two principal components was approximately 98%. Since 98% exceeded the preset 95% threshold, the eigenvectors corresponding to the two largest eigenvalues ​​(k=2) were selected. These two 4-dimensional eigenvectors together constitute a 4×2-dimensional projection matrix. The two-dimensional space spanned by this matrix is ​​the feature subspace capable of detecting process fluctuation information.

[0017] To assess the contribution of each process parameter to potential production failures across both long-term and short-term time scales, in an optional embodiment, calculating the composite failure contribution weight of each state parameter within the feature subspace includes: The spatiotemporal state matrix is ​​projected onto the feature subspace and then reconstructed to obtain the filtered spatiotemporal state matrix; For any state parameter in the filtered spatiotemporal state matrix, extract the corresponding value of the parameter in each row of the matrix to form a production time series vector; Calculate the standard deviation of the production time series vector as the statistical dispersion across batches; The energy spectral density of the production time series vector is calculated using Fast Fourier Transform, and the energy integral value within a preset abnormal fluctuation frequency band is calculated as a frequency domain fluctuation index. After normalizing the statistical dispersion and frequency domain fluctuation index respectively, the normalized statistical dispersion is multiplied by the first weighting coefficient, and the normalized frequency domain fluctuation index is multiplied by the second weighting coefficient. The sum of the two yields the composite fault contribution weight.

[0018] By projecting the original data onto principal component subspaces and then reconstructing it, data filtering effectively removes random noise interference while retaining the main process characteristics. This model structure incorporates two complementary evaluation dimensions: one to assess the dispersion of production parameters in amplitude, used for statistical stability; and the other to assess whether the production sequence exhibits abnormal periodic oscillations, used to measure frequency domain stability. These two dimensions are then weighted and fused.

[0019] Taking grouting pressure as an example, after projecting and reconstructing the 61×4-dimensional spatiotemporal state matrix, a filtered matrix is ​​obtained. The column vector corresponding to the grouting pressure in this matrix is ​​extracted, which contains a pressure value sequence for 61 production cycles, including previously qualified products, recently processed products, and currently awaiting processing products. The sample standard deviation of this 61-dimensional pressure sequence is calculated. If the value is 0.05 MPa, this value reflects the amplitude fluctuation of the grouting pressure in continuous production. Next, this 61-dimensional pressure sequence is treated as a discrete-time signal that varies with production batches, and a Fast Fourier Transform is performed to obtain the energy spectral density function. This function is integrated within a preset process fluctuation frequency band, for example, within a frequency range corresponding to a cyclic fluctuation occurring every 5-10 products, to obtain the energy integral value, such as 1.8, reflecting whether there is periodic pressure pulsation interference in the equipment. The statistical dispersion values ​​of all parameters and the energy integral value are then subjected to max-min normalization to ensure that the range falls between 0 and 1. Assuming the statistical dispersion of the normalized grouting pressure is 0.8, the energy integral is 0.6, and the first and second weighting coefficients are both 0.5, then the weight of the composite fault contribution is calculated to be 0.7. Figure 3 As shown.

[0020] In an improved embodiment, the calculation of the composite fault contribution weight also incorporates a mean deviation term. This involves calculating the absolute deviation between each state parameter of the current toilet to be processed and the mean values ​​of the corresponding parameters of the 10 preceding qualified products, and then normalizing this deviation. The composite fault contribution weight is obtained by weighted summation of the normalized statistical dispersion, frequency domain fluctuation index, and mean deviation term. This embodiment, by introducing mean deviation, overcomes the limitation of relying solely on variance and FFT to identify trend drift in process parameters.

[0021] In another optional embodiment, the calculation object of the frequency domain fluctuation index is limited to the continuous sampling sequence of the toilet to be processed within a single process cycle, thereby ensuring that the extracted energy features represent real disturbances such as grouting pump pulsation or pipeline vibration.

[0022] S3, perform tensor shrinking operation on the composite fault contribution weight and the preset fault feature basis vectors representing cracks, deformations and pinhole defects respectively to obtain a predicted fault vector containing multiple potential fault probabilities; based on the optimization model with the objective function of minimizing the norm of the predicted fault vector and the norm of the process control parameter adjustment amount, solve to obtain the initial process control command. Based on historical production data and failure analysis, a fault feature basis vector is established for three typical defects: cracking, deformation, and pinholes. The dimension of the vector is the same as the number of state parameters. For example, in the basis vector of cracking defects, the element values ​​related to the cooling rate of the firing process and the humidity of the drying process are relatively large; in the basis vector of deformation defects, the element values ​​related to the uniformity of grouting slurry density and the initial drying temperature are relatively large. The composite fault contribution weights of all state parameters are combined into a weight vector. This weight vector is then multiplied by the feature basis matrix formed by the three fault feature basis vectors. The result is a three-dimensional column vector, whose three elements correspond to the predicted probability values ​​of cracking, deformation, and pinhole defects, respectively, and together constitute the predicted fault vector.

[0023] An objective function is established, which is the sum of two terms. For example, the first term is the square of the L2 norm of the predicted fault vector multiplied by a weighting coefficient α, and the second term is the square of the L2 norm of the process control parameter adjustment vector multiplied by a weighting coefficient β. The process control parameter adjustment vector may include, for example, adjustments to grouting pressure, drying chamber target temperature, and firing kiln heating rate. By establishing a process model between process parameters and control parameters, the predicted fault vector is expressed as a function of process control parameter adjustments. Numerical optimization algorithms, such as gradient descent or sequential quadratic programming, are used to find the process control parameter adjustment vector that minimizes the objective function; this vector is the initial process control command.

[0024] In an optional embodiment, the step of performing tensor shrinking operation on the composite fault contribution weight and preset fault feature basis vectors representing cracks, deformations, and pinhole defects respectively to obtain a predicted fault vector containing multiple potential fault probabilities includes: The composite fault contribution weights of the four state parameters are used to construct a 4-dimensional weight vector. The three 4D fault feature basis vectors representing cracks, deformations, and pinholes are respectively used to construct a 3×4-dimensional fault feature basis matrix. The fault feature basis matrix and the weight vector are multiplied by matrix to obtain a 3-dimensional predicted fault vector. The three components of the predicted fault vector correspond to the probability of occurrence of cracks, deformations and pinholes, respectively.

[0025] Optionally, an abstract process parameter contribution weighting can be transformed into a prediction of the probability of specific product defects by using a linear mapping model. The core of the model is a pre-constructed fault feature basis matrix, which encodes the process expert's prior knowledge about how fluctuations in different parameters can trigger specific types of faults. This process involves a matrix-vector multiplication operation, i.e., a tensor contraction.

[0026] The composite fault contribution weights of the four parameters calculated above are combined into a 4-dimensional column vector. For example, the weight vector... These correspond to grouting pressure, mud temperature, grouting time, and ambient humidity, respectively. A pre-defined 3×4 dimensional fault feature basis matrix B is used, where each row represents a fault mode and each column corresponds to a process parameter. For example, the first row represents crack defects, with a vector of [0.9, 0.1, 0.2, 0.1], indicating that cracking is mainly related to the high contribution weight of grouting pressure. The second and third rows represent the feature basis vectors of deformation and pinhole defects, respectively. By calculating matrix multiplication P=BW, the 4-dimensional weight vector is mapped to a 3-dimensional predicted fault vector P. The first component of the calculation result represents the probability of cracking, the second represents the probability of deformation, and the third represents the probability of pinhole. For example, if the calculated... This indicates that the risk of deformation defects is highest under the current process conditions.

[0027] Because there is a nonlinear relationship between ceramic defects and process parameters, in an optional embodiment, the generation of the predicted fault vector uses a nonlinear mapping model instead of linear matrix operations. Specifically, a 4-dimensional weight vector is input into a pre-trained deep belief network or a multilayer perceptron with a Softmax output layer. The nonlinear activation function of the hidden layer is used to obtain nonlinear relationships such as the simultaneous abnormality of temperature and humidity inducing cracking, thereby improving the accuracy of the predicted fault vector under extreme process conditions.

[0028] To calculate the optimal process parameter adjustment scheme by solving an optimization problem, in an optional embodiment, the initial process control command is obtained by solving an optimization model with the objective function of minimizing the weighted sum of the norms of the predicted fault vector and the process control parameter adjustment amounts, including: Establish the objective function Where P(U) is the predicted fault vector under process control parameter U, This is a vector of adjustment values ​​for process control parameters. Denotes the square of the L2 norm. and These are weighting coefficients; The adjustment vector that minimizes the objective function J is obtained by iteratively solving using sequential quadratic programming. The initial process control command is obtained.

[0029] The established optimization model structure is a composite objective function with two objective terms. The first term... The goal is to minimize the failure risk of future products, that is, to make the overall magnitude of the predicted failure vector as small as possible. (Second term) The goal is to minimize control costs, that is, to avoid drastic adjustments to existing process parameters and ensure a smooth transition in the production process. Weighting coefficients and This reflects a strategy that prioritizes mitigating failure risks while also considering control stability. During the calculation, each quantity in the predicted failure vector is pre-normalized, and similarly, each quantity in the adjustment vector of the process control parameters is normalized.

[0030] Vector of process parameter adjustments As variables to be solved, based on historical data, a correlation model between process parameters and state parameters / failure rates is constructed using partial least squares (PLS) or neural networks. Since the relationship between the predicted fault vector P(U) and the process parameter U is non-linear, a sequential quadratic programming method is used. In each iteration, the objective function near the current point is approximated by a quadratic function, and the constraints are linearized, thus transforming the original problem into an easily solvable quadratic programming subproblem. By repeatedly solving this subproblem and updating the adjustment vector... The algorithm gradually approaches the optimal point that minimizes the objective function J. When the algorithm converges, the resulting optimal adjustment vector is... For example, [reduce by 0.2 MPa, increase by 30 seconds, reduce by 1.2 degrees Celsius, increase drying wind speed by 0.1 m / s] is the initial process control command.

[0031] S4, When the adjustment amount of any parameter in the initial process control command exceeds the threshold, apply an asymmetric penalty term related to the adjustment direction to the command and solve it again to obtain the process control command.

[0032] Calculate the L2 norm of the predicted fault vector. The larger the norm, the higher the potential fault risk, and the smaller the threshold. The calculation formula is, for example: Threshold = Constant C / L2 norm of the predicted fault vector. Iterate through each adjustment in the initial process control command, such as increasing the drying temperature by 2°C. If the absolute value is greater than the calculated threshold, a penalty is triggered. At this point, an asymmetric penalty term is added to the original objective function. For example, if increasing the temperature results in greater equipment loss or risk than decreasing it, the penalty term can be set as a piecewise function, with a larger penalty when the adjustment is positive and a smaller penalty when it is negative. The strength coefficient of the penalty term is proportional to the correlation degree, which is queried from a preset knowledge base matrix. For example, if the deformation probability is the highest in the predicted fault vector, and the current adjustment parameter exceeding the threshold is the drying temperature, the knowledge base shows that the correlation degree between the drying process and deformation defects is 0.8. This correlation degree is then multiplied by a basic penalty coefficient to obtain the penalty strength. The new objective function with the penalty term is then optimized again, and the result is the process control command.

[0033] To establish a judgment criterion, in an optional embodiment, the threshold is calculated using the following formula: ,in, For the threshold, Given the current predicted fault vector, C is the base threshold constant. It is a tiny positive number used to prevent the denominator from being zero.

[0034] Specifically, when in a high-fault-risk state, the tolerance for intervention in the control system is higher, therefore the threshold should be lower; conversely, when operation is smooth and the fault risk is very low, the threshold should be higher to avoid unnecessary adjustments and alarms. Obtain the predicted fault vector under the current state. For example, in the vector obtained above, Calculate the L2 norm of the vector, which is the square root of the sum of the squares of its components. The norm value represents the current overall failure risk level. This norm value is substituted into the formula for calculation, where the basic threshold constant C is set to 0.5, and the small normal number... Set it to 0.001. Threshold. The calculated result is approximately 0.403. This threshold will be compared with the norm of the calculated control adjustment amount. If the adjustment norm exceeds 0.403, a manual confirmation request will be triggered. Figure 4 .

[0035] In an optional embodiment, applying an asymmetric penalty term related to the adjustment direction to the instruction and resolving the problem includes: For the adjustment vector any component If the increase is positive, then a penalty term is added to the objective function. ; If the component decreases negatively, a penalty term is added to the objective function. ; in , The asymmetric penalty coefficient is used. The correlation value between the process affected by the pre-calibrated i-th parameter and the current maximum probability fault type is [0,1]. The objective function with a penalty term is re-optimized and solved to obtain process control instructions.

[0036] The initial optimization model is refined using an asymmetric penalty term incorporating knowledge of the process mechanism, thereby aligning control commands with actual production constraints and costs. This model imposes different penalty strengths on parameter increases and decreases because, in actual production, increasing temperature or pressure typically leads to higher equipment damage or safety risks than decreasing them. Furthermore, the penalty magnitude is also related to the correlation between the parameter and the current fault type. Related. Optionally, , Identify the most probable fault type, such as deformation in the example above. Query the correlation values ​​between various process parameters and deformation faults from the expert knowledge base. Assuming that temperature has the strongest correlation with deformation, However, the correlation with grouting pressure is low. The new penalty term is added to the original objective function J. For example, it is used to adjust the drying temperature. If the optimizer attempts to increase the temperature, the objective function will be incremented by a value. The penalty; if you try to lower the temperature, add a value... The penalty is increased. Because the penalty coefficient is larger, the optimizer will tend to avoid increasing the temperature. The new objective function with the added asymmetric penalty is re-optimized using sequential quadratic programming. The resulting new adjustment vector is the process control instruction that considers the process mechanism and adjustment cost, such as... Figure 5 In an optional embodiment, the coefficient of the asymmetric penalty term is also related to the current ambient temperature, humidity, and the kiln's service life. For example, towards the end of the kiln's operation, the penalty coefficient for adjusting the firing temperature is increased. Automatic adjustment is implemented to prevent excessive wear of kiln refractory materials, thereby achieving synergistic optimization of process quality and equipment lifespan.

[0037] In the second embodiment, the present invention also proposes a manufacturing process control system for split toilets, comprising the following modules: The module is used to obtain the real-time status parameters of the current split toilet to be processed in each process, and to construct a spatiotemporal state matrix by combining the historical status parameter sets of multiple previous qualified products and recently processed toilets. The calculation module is used to perform eigenvalue decomposition and dimensionality reduction on the spatiotemporal state matrix to construct a feature subspace containing the main process fluctuation information; within the feature subspace, the composite fault contribution weight of each state parameter is calculated, wherein the weight is positively correlated with the statistical dispersion of the state parameter across batches and the energy spectral density integral value in the current batch; The first solution module is used to perform tensor shrinking operation on the composite fault contribution weight and the preset fault feature basis vectors representing cracks, deformations and pinhole defects respectively, to obtain a predicted fault vector containing multiple potential fault probabilities; and to obtain the initial process control command based on an optimization model with the objective function of minimizing the norm of the predicted fault vector and the norm of the process control parameter adjustment amount. The second solution module is used to apply an asymmetric penalty term related to the adjustment direction to the instruction and re-solve when the adjustment amount of any parameter in the initial process control instruction exceeds the threshold, so as to obtain the process control instruction.

[0038] In an optional embodiment, the step of obtaining the real-time status parameters of the current split toilet to be processed at each stage, and constructing a spatiotemporal state matrix by combining the historical status parameter sets of multiple preceding qualified products and recently processed toilets, includes: The four real-time status parameters of the split toilet to be processed are: grouting pressure, slurry temperature, slurry intake time, and ambient humidity. These parameters are used as one row of a matrix. Select the four status parameters of the first 10 qualified products in the corresponding process as 10 rows; We selected 50 recently processed toilets and used 4 state parameters of the corresponding process as 50 rows to form a 61×4-dimensional spatiotemporal state matrix.

[0039] In an optional embodiment, the step of performing eigenvalue decomposition and dimensionality reduction on the spatiotemporal state matrix to construct a feature subspace containing information on the main process fluctuations includes: The spatiotemporal state matrix is ​​processed using principal component analysis. Calculate the covariance matrix of the matrix, and solve for the eigenvalues ​​and eigenvectors of the covariance matrix; The eigenvectors corresponding to the k largest eigenvalues ​​with a cumulative variance contribution rate of 95% are selected to form the feature subspace.

[0040] In an optional embodiment, calculating the composite fault contribution weight of each state parameter within the feature subspace includes: The spatiotemporal state matrix is ​​projected onto the feature subspace and then reconstructed to obtain the filtered spatiotemporal state matrix; For any state parameter in the filtered spatiotemporal state matrix, extract the corresponding value of the parameter in each row of the matrix to form a production time series vector; Calculate the standard deviation of the production time series vector as the statistical dispersion across batches; The energy spectral density of the production time series vector is calculated using Fast Fourier Transform, and the energy integral value within a preset abnormal fluctuation frequency band is calculated as a frequency domain fluctuation index. After normalizing the statistical dispersion and frequency domain fluctuation index respectively, the normalized statistical dispersion is multiplied by the first weighting coefficient, and the normalized frequency domain fluctuation index is multiplied by the second weighting coefficient. The sum of the two yields the composite fault contribution weight.

[0041] In an optional embodiment, the step of performing tensor shrinking operation on the composite fault contribution weight and preset fault feature basis vectors representing cracks, deformations, and pinhole defects respectively to obtain a predicted fault vector containing multiple potential fault probabilities includes: The composite fault contribution weights of the four state parameters are used to construct a 4-dimensional weight vector. The three 4D fault feature basis vectors representing cracks, deformations, and pinholes are respectively used to construct a 3×4-dimensional fault feature basis matrix. The fault feature basis matrix and the weight vector are multiplied by matrix to obtain a 3-dimensional predicted fault vector. The three components of the predicted fault vector correspond to the probability of occurrence of cracks, deformations and pinholes, respectively.

[0042] In an optional embodiment, the initial process control command is obtained by solving an optimization model based on a weighted sum of the norms of the predicted fault vector and the process control parameter adjustment amounts as the objective function, including: Establish the objective function Where P(U) is the predicted fault vector under process control parameter U, This is a vector of adjustment values ​​for process control parameters. Denotes the square of the L2 norm. and These are weighting coefficients; The adjustment vector that minimizes the objective function J is obtained by iteratively solving using sequential quadratic programming. The initial process control command is obtained.

[0043] In an optional embodiment, the threshold is calculated using the following formula: ,in, For the threshold, Given the current predicted fault vector, C is the base threshold constant. It is a tiny positive number used to prevent the denominator from being zero.

[0044] In an optional embodiment, applying an asymmetric penalty term related to the adjustment direction to the instruction and resolving the problem includes: For the adjustment vector any component If the increase is positive, then a penalty term is added to the objective function. ; If the component decreases negatively, a penalty term is added to the objective function. ; in , The asymmetric penalty coefficient is used. The correlation value between the process affected by the pre-calibrated i-th parameter and the current maximum probability fault type is [0,1]. The objective function with a penalty term is re-optimized and solved to obtain process control instructions.

[0045] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0046] The functional modules shown in the above-described block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0047] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0048] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0049] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for controlling the manufacturing process of a split toilet, characterized in that, Includes the following steps: The system acquires the real-time status parameters of the split toilet to be processed at each stage, and combines the historical status parameter sets of multiple qualified products from the previous stage and recently processed toilets to construct a spatiotemporal state matrix. The spatiotemporal state matrix is ​​subjected to eigenvalue decomposition and dimensionality reduction to construct a feature subspace containing information on the main process fluctuations; within the feature subspace, the composite fault contribution weight of each state parameter is calculated. The composite fault contribution weight is combined with the preset fault feature basis vectors representing cracks, deformations and pinholes respectively by tensor shrinkage operation to obtain a predicted fault vector containing multiple potential fault probabilities; the initial process control command is obtained by solving an optimization model with the objective function of minimizing the norm of the predicted fault vector and the norm of the process control parameter adjustment. When the adjustment amount of any parameter in the initial process control command exceeds the threshold, an asymmetric penalty term related to the adjustment direction is applied to the command and the solution is re-solved to obtain the process control command.

2. The method according to claim 1, characterized in that, The process involves acquiring the real-time status parameters of the current split-type toilet under processing at each stage, and combining these parameters with the historical status parameter sets of multiple previously qualified products and recently processed toilets to construct a spatiotemporal state matrix, including: The four real-time status parameters of the split toilet to be processed are: grouting pressure, slurry temperature, slurry intake time, and ambient humidity. These parameters are used as one row of a matrix. Select the four status parameters of the first 10 qualified products in the corresponding process as 10 rows; We selected 50 recently processed toilets and used 4 state parameters of the corresponding process as 50 rows to form a 61×4-dimensional spatiotemporal state matrix.

3. The method according to claim 1, characterized in that, The step of performing eigenvalue decomposition and dimensionality reduction on the spatiotemporal state matrix to construct a feature subspace containing information on the main process fluctuations includes: The spatiotemporal state matrix is ​​processed using principal component analysis. Calculate the covariance matrix of the matrix, and solve for the eigenvalues ​​and eigenvectors of the covariance matrix; The eigenvectors corresponding to the k largest eigenvalues ​​with a cumulative variance contribution rate of 95% are selected to form the feature subspace.

4. The method according to claim 1, characterized in that, The calculation of the composite fault contribution weight of each state parameter within the feature subspace includes: The spatiotemporal state matrix is ​​projected onto the feature subspace and then reconstructed to obtain the filtered spatiotemporal state matrix; For any state parameter in the filtered spatiotemporal state matrix, extract the corresponding value of the parameter in each row of the matrix to form a production time series vector; Calculate the standard deviation of the production time series vector as the statistical dispersion across batches; The energy spectral density of the production time series vector is calculated using Fast Fourier Transform, and the energy integral value within a preset abnormal fluctuation frequency band is calculated as a frequency domain fluctuation index. After normalizing the statistical dispersion and frequency domain fluctuation index respectively, the normalized statistical dispersion is multiplied by the first weighting coefficient, and the normalized frequency domain fluctuation index is multiplied by the second weighting coefficient. The sum of the two yields the composite fault contribution weight.

5. The method according to claim 1, characterized in that, The step involves performing a tensor shrinking operation on the composite fault contribution weights and preset fault feature basis vectors representing cracks, deformations, and pinhole defects, respectively, to obtain a predicted fault vector containing multiple potential fault probabilities, including: The combined fault contribution weights of the four state parameters are used to construct a 4-dimensional weight vector. The three 4D fault feature basis vectors representing cracks, deformations, and pinholes are respectively used to construct a 3×4-dimensional fault feature basis matrix. The fault feature basis matrix and the weight vector are multiplied by matrix to obtain a 3-dimensional predicted fault vector. The three components of the predicted fault vector correspond to the probability of occurrence of cracks, deformations and pinholes, respectively.

6. The method according to claim 1, characterized in that, The initial process control command is obtained by solving an optimization model based on a weighted sum of the norms of the predicted fault vector and the process control parameter adjustment amounts as the objective function, including: Establish objective function Where P(U) is the predicted fault vector under process control parameter U, This is a vector of adjustment values ​​for process control parameters. Denotes the square of the L2 norm. and These are weighting coefficients; The adjustment vector that minimizes the objective function J is obtained by iteratively solving using sequential quadratic programming. The initial process control command is obtained.

7. The method according to claim 1, characterized in that, The formula for calculating the threshold is: ,in, For the threshold, Given the current predicted fault vector, C is the base threshold constant. It is a tiny positive number used to prevent the denominator from being zero.

8. The method according to claim 1, characterized in that, The step of applying an asymmetric penalty term related to the adjustment direction to the instruction and resolving it includes: For the adjustment vector any component If the increase is positive, then a penalty term is added to the objective function. ; If the component decreases negatively, a penalty term is added to the objective function. ; in , The asymmetric penalty coefficient is used. This is the correlation value between the process affected by the pre-calibrated i-th parameter and the current maximum probability fault type, with a value range of [0,1]. The objective function with a penalty term is re-optimized and solved to obtain process control instructions.

9. A manufacturing process control system for a split-type toilet, characterized in that, Includes the following modules: The module is used to obtain the real-time status parameters of the current split toilet to be processed in each process, and to construct a spatiotemporal state matrix by combining the historical status parameter sets of multiple previous qualified products and recently processed toilets. The calculation module is used to perform eigenvalue decomposition and dimensionality reduction on the spatiotemporal state matrix to construct a feature subspace containing information on the main process fluctuations; within the feature subspace, the composite fault contribution weight of each state parameter is calculated. The first solution module is used to perform tensor shrinking operation on the composite fault contribution weight and the preset fault feature basis vectors representing cracks, deformations and pinhole defects respectively, to obtain a predicted fault vector containing multiple potential fault probabilities; and to obtain the initial process control command based on an optimization model with the objective function of minimizing the norm of the predicted fault vector and the norm of the process control parameter adjustment amount. The second solution module is used to apply an asymmetric penalty term related to the adjustment direction to the instruction and re-solve when the adjustment amount of any parameter in the initial process control instruction exceeds the threshold, so as to obtain the process control instruction.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when executed by a processor, implements the method as described in any one of claims 1-8.

Images