A system and method for testing the mixing uniformity of a stirred tank

Abstract

The application relates to the technical field of food production and process analysis, and discloses a stirring tank mixing uniformity test system and method, which comprises the following modules: a space-time light field excitation module for projecting a preset space-time light field to materials in a stirring tank; a high-dimensional light field response capture module for capturing response light generated by the materials in response to the space-time light field and generating response data; a model storage module for storing a forward physical model representing the concentration distribution relationship between the response data and key components of the materials; an inversion calculation module for calculating the spatial covariance matrix of the concentration distribution of the key components based on the response data and the forward physical model; and an index generation and display module for generating and outputting a quantitative index representing the mixing uniformity of the materials based on the spatial covariance matrix. The application realizes non-invasive, real-time and quantitative monitoring of the global mixing uniformity, and improves the control efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of food production and process analysis, and in particular to a system and method for testing mixing uniformity of a stirring tank. Background Art

[0002] In food production, mixing tanks are key equipment for achieving material mixing. Mixing uniformity is a key parameter that determines the quality of the final product. Accurately determining the mixing endpoint is particularly crucial when producing solid, liquid, and semi-solid compound seasonings and food flavors (liquids).

[0003] Currently, the primary method for determining blend uniformity relies on offline sampling and analysis, whereby the equipment is paused and samples are taken for testing. Alternatively, there are technologies that utilize online probe sensors, such as conductivity or spectroscopy probes, that are inserted directly into the material for point-of-care measurement.

[0004] However, these technologies have significant shortcomings. Offline sampling and analysis are subject to time delays, making real-time control difficult and the sampling representativeness poor. While online probes can measure in real time, their invasive installation interferes with the flow field within the mixing tank, affecting processes such as mixing and emulsification. They are also susceptible to contamination when processing high-viscosity materials such as slurry (paste) seasonings. More importantly, both provide only localized information and lack a means to globally assess the material's distribution status.

[0005] Therefore, the present invention proposes a mixing uniformity testing system and method for a stirring tank to address the deficiencies of the prior art. Summary of the Invention

[0006] The purpose of the present invention is to provide a stirring tank mixing uniformity testing system and method, which solves the problem that the existing technology is difficult to perform real-time, quantitative, non-invasive and global evaluation of the mixing uniformity of the stirring tank.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A mixing uniformity testing system for a stirring tank, comprising:

[0008] The spatiotemporal light field excitation module is used to project a preset spatiotemporal light field onto the material in the mixing tank;

[0009] a high-dimensional light field response capture module configured to capture the response light generated by the material in the mixing tank to the spatiotemporal light field and generate response data;

[0010] A model storage module, configured to store a preset forward physical model, wherein the forward physical model is configured to characterize the relationship between the response data and the concentration distribution of key components within the material in the mixing tank;

[0011] an inversion calculation module configured to calculate a spatial covariance matrix of the concentration distribution of the key component based on the response data and the forward physical model by solving an inverse problem;

[0012] an index generation and display module configured to receive the spatial covariance matrix and generate and output at least one quantitative index for characterizing the mixing uniformity of the material in the stirred tank based on the spatial covariance matrix.

[0013] Preferably, the spatiotemporal light field excitation module comprises:

[0014] at least one light source configured to generate a modulated probe light beam;

[0015] at least one scanning device configured to control the probe light beam to project onto the material in the stirred tank according to a preset scanning trajectory, thereby forming the preset spatiotemporal light field.

[0016] Preferably, the high-dimensional light field response capturing module comprises:

[0017] receiving the response light generated by the material in the stirred tank from multiple different positions on the surface of the material;

[0018] converting the received response light into one or more electrical signals;

[0019] processing the one or more electrical signals to generate the response data.

[0020] Preferably, the forward physical model in the model storage module is established based on radiative transfer theory describing the propagation process of photons in the material in the stirred tank as a scattering medium.

[0021] Preferably, the forward physical model utilizes a sensitivity matrix The relationship between the response data and the internal concentration distribution of the key component in the material in the stirred tank is defined as the following mathematical expression:

[0022] ;

[0023] wherein, is a vector composed of the response data generated by the high-dimensional light field response capturing module; is the sensitivity matrix corresponding to the forward physical model; is a vector representing the concentration distribution of the key component; is a measurement noise vector contained in the response data.

[0024] Preferably, the inversion calculation module comprises:

[0025] solving the inverse problem based on the response data and the forward physical model:

[0026] constructing a likelihood function based on the forward physical model and a measurement noise characteristic in the response data;

[0027] setting a prior probability distribution for the concentration distribution of the key component;

[0028] calculating, under a Bayesian inference framework, in combination with the likelihood function and the prior probability distribution, to obtain the spatial covariance matrix.

[0029] Preferably, the calculating, under a Bayesian inference framework, in combination with the likelihood function and the prior probability distribution, to obtain the spatial covariance matrix comprises:

[0030] determining a posterior probability distribution describing the concentration distribution of the key component according to the Bayesian inference framework defined by Bayes' theorem wherein the mathematical relationship of the Bayes' theorem is:

[0031]

[0032] wherein, is the posterior probability distribution; is the likelihood function; is the prior probability distribution;

[0033] calculating a covariance of the posterior probability distribution to obtain the spatial covariance matrix based on the determined posterior probability distribution, wherein the calculating is performed by solving a posterior covariance matrix formula:

[0034]

[0035] wherein, is an estimation of the spatial covariance matrix; is a sensitivity matrix corresponding to the forward physical model; is a transpose of the sensitivity matrix; is a covariance matrix of measurement noise; is a prior covariance matrix of the concentration distribution of the key component.

[0036] Preferably, the index generating and displaying module comprises:

[0037] receiving the spatial covariance matrix calculated by the inversion calculation module;

[0038] ​​Based on the received spatial covariance matrix, a numerical scalar is calculated and at least one quantitative index for representing the mixing uniformity of the material in the stirred tank is output, wherein the quantitative index includes at least one of the following group:

[0039] The trace of the spatial covariance matrix;

[0040] The determinant of the spatial covariance matrix;

[0041] The matrix norm of the spatial covariance matrix;

[0042] Display the generated quantitative index.

[0043] The present application also provides a mixing uniformity testing method of a stirred tank, comprising the following steps:

[0044] Projecting a preset spatio-temporal light field to the material in the stirred tank;

[0045] Capturing the response light generated by the material in the stirred tank to the spatio-temporal light field and generating response data;

[0046] Storing a preset forward physical model, which is used to represent the relationship between the response data and the internal key component concentration distribution of the material in the stirred tank;

[0047] Based on the response data and the forward physical model, the spatial covariance matrix of the key component concentration distribution is calculated by solving the inverse problem;

[0048] Receiving the calculated spatial covariance matrix and generating and outputting at least one quantitative index for representing the mixing uniformity of the material in the stirred tank based on the spatial covariance matrix.

[0049] In summary, the present application includes at least one of the following beneficial technical effects:

[0050] 1. The present application obtains response data in a non-invasive manner through a spatio-temporal light field excitation module and a high-dimensional light field response capture module, and combines inversion calculation and index generation process to realize online quantitative monitoring of the mixing uniformity of the material in the stirred tank. This scheme avoids the hysteresis and destructiveness of traditional sampling analysis, and also overcomes the problem of flow field interference caused by the insertion probe, and can provide continuous quantitative evaluation basis for the production process of solid, liquid, semi-solid composite seasoning and food essence (liquid) products.

[0051] 2.The application provides a reliable physical basis for the measurement of complex materials by using a forward physical model based on radiation transfer theory and performing inversion calculation using a Bayesian inference framework. This scheme does not rely on empirical fitting, and by fusing measurement noise and prior information in the algorithm, it effectively suppresses the influence of measurement error on the results and improves the robustness of the calculation results. This makes the application applicable to the mixing uniformity testing of industrial materials such as composite seasonings with complex and variable optical properties, ensuring the accuracy of the measurement results.

[0052] 3.The application directly solves the spatial covariance matrix in the inversion calculation, rather than reconstructing the complete concentration distribution image, thereby avoiding the problems of large calculation amount and instability caused by solving ill-posed inverse problems. This method combines the single quantitative index output by the index generation module, reduces the calculation load while ensuring the stability of the results, and meets the real-time requirements of online monitoring. The index curve output by the method is intuitive and convenient for operators to judge the mixing end point of process steps such as'mixing' of powder seasonings, 'emulsification stirring' and 'homogenization' of emulsified essences, and 'heating stirring' of paste (cream) seasonings, providing a clear basis for process control. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 Fig. 1 is a schematic diagram of the system architecture of the application;

[0054] Figure 2 Fig. 2 is a schematic diagram of the method flow of the application. DETAILED DESCRIPTION

[0055] The following will be described in detail with reference to the accompanying drawings Figure 1 - the accompanying drawings Figure 2 , to further illustrate the application.

[0056] The embodiment of the application provides a stirring tank mixing uniformity testing system, which comprises:

[0057] A spatiotemporal light field excitation module is used to project a preset spatiotemporal light field to the material in the stirring tank.

[0058] In this embodiment, the spatiotemporal light field excitation module serves as the excitation signal source of the entire testing system of the application, and its core function is to actively perform optical detection on the material in the stirring tank to provide high-quality original excitation signals carrying the internal state information of the material for subsequent response capture and inversion calculation.

[0059] The design of the spatiotemporal light field excitation module aims to solve the industrial mixing process, especially in the production process of solid, liquid and semi-solid composite seasonings, and to monitor the mixing state of materials with complex optical properties such as slurry, suspension or powder. In these scenarios, due to the high scattering and high absorption characteristics of the materials, it is difficult to obtain comprehensive internal structure information using traditional single-point, static measurement methods.

[0060] Therefore, the present application creatively proposes to use "spatiotemporal light field" for active detection. The technical advantage is that it can more fully and stereoscopically explore the three-dimensional space in the stirring tank. Specifically, the controllability of the "spatial" dimension, i.e. allowing the detection light to enter the material from different positions and angles, ensures that the optical signal can explore a wider volume of material, avoiding the deviation caused by local measurement; and the controllability of the "time" dimension, i.e. the modulation of the light beam, aims to accurately distinguish the effective detection signal from the complex industrial environment background light, laying a solid foundation for subsequent high signal-to-noise ratio measurement.

[0061] Further, in order to achieve the above functions, the spatiotemporal light field excitation module, in a preferred embodiment, has the following specific composition and working mode:

[0062] The spatiotemporal light field excitation module includes at least one light source for generating a modulated detection light beam. Preferably, the light source can use one or more semiconductor laser diodes (Laser Diode) or high-intensity light-emitting diodes (LED). When dealing with materials such as the aforementioned sauces and sauces that are rich in organic matter and water, the emission wavelength of the light source can be preferably set in the near-infrared (NIR) band, such as a specific wavelength in the range of 700 nm to 1000 nm. The reason for using this band is that the photons have a relatively good tissue penetration depth in the above medium, and have distinguishable absorption or scattering characteristics for changes in the concentration of key components such as fat, protein and water, which is beneficial for information acquisition.

[0063] It is particularly important that the detection light beam is modulated. This means that the light beam emitted by the light source is not continuous light with constant intensity, but the light intensity changes periodically with high frequency over time, for example, the driving current of the light source can be modulated with a sine wave or square wave. The core purpose of this design is to provide the possibility for subsequent signal receiving end to demodulate using lock-in amplification and other special signal processing techniques. Through this technology, the extremely weak response light signal carrying material information can be accurately separated from the strong background light (such as workshop lighting, equipment heat radiation, etc.) noise in the industrial environment of the stirring tank, which is a key technical means to ensure the measurement accuracy and stability of the entire system.

[0064] In addition, the spatiotemporal light field excitation module further comprises at least one scanning device. The scanning device is configured to receive the modulated probe light beam generated by the light source and accurately control the light beam to project onto the outer wall of the stirring tank or an optical observation window arranged thereon according to a preset scanning trajectory. In a specific embodiment, the scanning device can be a high-speed two-axis galvanometer scanner or a digital micro-mirror device (DMD) capable of realizing high-precision and rapid deflection of the light beam.

[0065] The preset scanning trajectory herein refers to a spatial point array composed of dozens or even hundreds of discrete points, which are uniformly distributed in a rectangular or circular region of the tank body. During system operation, the scanning device drives the probe light beam to perform high-speed and cyclic point-by-point scanning according to the trajectory point array, temporarily stays at each point for a millisecond, and then rapidly switches to the next point.

[0066] Therefore, by combining the time-modulated probe light beam with the spatially scanned projection points, the spatiotemporal light field excitation module finally projects a structured spatiotemporal light field, which is accurately controlled in time and space and can be repeated, onto the material in the stirring tank.

[0067] The high-dimensional light field response capturing module is configured to capture the response light generated by the material in the stirring tank in response to the spatiotemporal light field and generate response data.

[0068] In this embodiment, the high-dimensional light field response capturing module is the sensing and measuring unit of the entire test system of the present application, which closely connects the aforementioned spatiotemporal light field excitation module in function and is responsible for accurately "listening to" and "recording" the response light signal generated by the material.

[0069] The core task of the high-dimensional light field response capturing module is to capture the response light emitted from the surface of the material after the probe light beam projected by the spatiotemporal light field excitation module undergoes complex optical effects such as absorption and scattering with the material in the tank (e.g., a composite seasoning material being mixed or stirred, an emulsified essence, or other solid-liquid mixed materials). This response light is no longer a simple original light beam, but a diffuse light field that has undergone significant changes in space and intensity, and its characteristics deeply contain rich information about the spatial distribution state of the internal material, especially the key components. The design purpose of this module is to capture this information in a high-fidelity and high-dimensional manner, providing the original basis for subsequent quantitative analysis.

[0070] Further, in order to comprehensively and with high signal-to-noise ratio capture the response light and ultimately generate structured data required for subsequent calculation, the high-dimensional light field response capturing module has the following specific working process and structure in a preferred embodiment:

[0071] The high-dimensional light field response capturing module first receives the response light generated by the material to the spatiotemporal light field from multiple different positions on the surface of the material in the stirring tank. To achieve this function, a detector array composed of multiple independent high-sensitivity photodetectors is preferably arranged on the outer wall of the stirring tank around the area excited by the aforementioned spatiotemporal light field. The geometric configuration of the array can be annular, rectangular, or other suitable arrangements. The technical concept of using a detector array instead of a single detector is that the dispersed response light field can be sampled from multiple spatial perspectives and positions simultaneously to obtain its spatial distribution characteristics. This multi-point parallel detection is the physical basis for constructing "high-dimensional" response data, which is crucial for subsequent accurate inversion of the internal structure.

[0072] Preferably, the photodetectors can be selected from avalanche photodiodes (APD) or single-photon avalanche diodes (SPAD) and the like. The reason for selecting such devices is that they have extremely high detection efficiency and fast response capability for the weak light signals in the aforementioned near-infrared band, and are very suitable for capturing the effective signal that has become very attenuated after penetrating the highly scattering and highly light-absorbing food material.

[0073] After receiving the response light, the high-dimensional light field response capturing module converts the received response light into one or more electrical signals. Specifically, each detector unit in the array generates an analog electrical signal, such as a weak current or voltage signal, with an intensity that is a specific function of the intensity of the incident light, according to its internal photoelectric conversion effect, after receiving the response photons at its corresponding position. At this time, the entire detector array outputs a set of parallel raw electrical signals representing the spatial intensity distribution of the response light.

[0074] Subsequently, the module processes the aforementioned one or more electrical signals to generate response data. This is a multi-step signal conditioning and digitization process that is crucial for ensuring data quality. The analog electrical signals carrying information are first sent to a multi-channel data processing and acquisition system. In this system, the signals are pre-amplified to boost their amplitude to the optimal processing range of the subsequent circuit; and then pass through a band-pass filter to filter out electronic noise and part of the environmental background light interference that is unrelated to the modulation frequency.

[0075] Especially crucial is that if the aforementioned spatiotemporal light field excitation module uses intensity modulation technology, a corresponding demodulation step will be included in this processing step. For example, a lock-in amplification technique can be used, which uses a reference signal with the same frequency and phase as the light source modulation signal, to process the electrical signals output by the detectors. This demodulation process can extremely effectively extract the weak target signal submerged in a strong noise background, and is one of the core technical guarantees for the high-precision online measurement of the present application.

[0076] The analog signal after the above amplification, filtering, demodulation and other series of processing will be digitized by a high-speed analog-to-digital converter (ADC). The digitized readings of all detector channels at a specific time point collectively constitute a high-dimensional vector, i.e. the response data vector in the subsequent calculation steps of the present application Each element of this vector corresponds to a specific "light source incident position - detector receiving position" pair of measurement values, which completely and structurally records the "optical fingerprint" of the material at that moment.

[0077] The model storage module is used to store a preset forward physical model, which is used to characterize the relationship between the response data and the internal key component concentration distribution of the material in the stirring tank;

[0078] In this embodiment, the model storage module is specially used to store a preset forward physical model that can accurately describe the physical process. The function of the model storage module is to establish a certain and robust mathematical bridge between the external optical response data measured by the aforementioned high-dimensional light field response capture module and the internal key component concentration distribution of the material that is invisible inside the stirring tank and that we are really interested in.

[0079] The fundamental purpose of the model storage module is to provide a solid physical basis for subsequent inversion calculations. When dealing with slurries, sauces or liquids containing suspended particles with extremely complex optical properties encountered in the production process of composite seasonings, the relationship between the externally measured light signal and the internal microscopic concentration distribution is nonlinear and nonlocal, which is difficult to describe by simple empirical formulas or linear interpolation. Therefore, it is necessary to rely on a physical model based on first principles to accurately infer from "appearance" to "essence".

[0080] Further, the preset forward physical model stored in the model storage module is established in accordance with rigorous scientific theories and is expressed in an efficient mathematical form to meet the needs of real-time calculation. The specific technical connotation is as follows:

[0081] The forward physical model is established based on the radiative transfer theory (RTT) describing the propagation process of photons in the material in the stirring tank as a scattering medium. The radiative transfer theory is the gold standard theory for the propagation of light in a participating medium (i.e. a medium with both absorption and scattering). This theory is chosen as the basis of the model because it can accurately depict every absorption and scattering event experienced by the detected photons after entering highly scattering and highly absorbing materials such as seasoning sauce and emulsion, thereby accurately predicting the energy distribution of photons inside the medium and the final state when escaping from the medium.

[0082] In practical applications, analytical solutions of the radiative transfer equation are usually difficult to obtain for complex systems with specific geometric boundaries, such as stirred tanks. Therefore, the forward physical model in the present invention is preferably pre-computed by numerical simulation methods. For example, the Monte Carlo simulation method can be employed to statistically solve the optical transfer relationship between the light source-detector and the internal points of the material by tracking the propagation paths of hundreds of millions of virtual photons in a given stirred tank model (containing parameters such as its geometric dimensions, average absorption and scattering coefficients of the material, etc.).

[0083] In order to facilitate subsequent rapid calculation, this complex model based on radiative transfer theory is finally linearized and expressed as a specific mathematical expression. Specifically, the forward physical model uses a sensitivity matrix to define the relationship between the response data and the internal key component concentration distribution of the material in the stirred tank as the following mathematical expression:

[0084] ;

[0085] In the formula, is a vector composed of response data generated by the high-dimensional light field response capture module. This vector is a collection of signal values collected by all detector channels in a complete measurement period and after preprocessing. Its dimension is equal to (number of light source scanning points) multiplied by (number of detectors), and it is a high-dimensional measurement result containing rich spatiotemporal information.

[0086] is a vector representing the key component concentration distribution. This is the unknown quantity that we want to know but is difficult to measure directly. In modeling, the internal space of the stirred tank has been discretized into hundreds of thousands of tiny three-dimensional grid elements, i.e., voxels. Each element in the vector represents the disturbance value of the concentration of key components such as fat balls and solid seasoning particles relative to the background concentration in the th voxel.

[0087] is the measurement noise vector contained in the response data. It represents the sum of various random errors that are inevitable in the actual measurement process, mainly including electronic noise generated by the detector itself and shot noise inherent in the optical signal itself. Accurate estimation of the statistical properties (such as its covariance matrix) of this noise is crucial for subsequent high-quality inversion calculation.

[0088] is the sensitivity matrix corresponding to the forward physical model. This is the core of the entire forward physical model. This matrix is ​​a huge constant matrix that has been calculated through the aforementioned numerical simulation before the system is put into operation. Each element has a clear physical meaning: it quantifies the When the concentration of the internal voxels changes by a unit, How much response signal change is caused on a measurement channel (i.e. a specific light source-detector pair). Therefore, the sensitivity matrix All physically transmitted information, from internal material status to external optical measurement, is fully encoded.

[0089] The model storage module stores the pre-calculated, high-precision sensitivity matrix in its internal storage area (such as non-volatile memory). , providing a stable, reliable and quickly callable "physical law lookup table" for the entire system.

[0090] The inversion calculation module is used to calculate the spatial covariance matrix of the concentration distribution of key components by solving the inverse problem based on the response data and the forward physical model;

[0091] In this embodiment, the function of the inversion calculation module is based on the response data generated in real time by the high-dimensional light field response capture module, and calls the preset forward physical model stored in the model storage module to solve a mathematical physics inverse problem to ultimately calculate the key quantitative indicators that can directly characterize the material mixing state.

[0092] The inversion calculation module aims to solve the core technical challenge of inferring the complex, invisible internal physical state from limited, indirect external measurement data. In the field of optical tomography, this inverse problem is often ill-posed. Directly solving a complete, high-resolution three-dimensional image of the internal concentration distribution is not only computationally intensive and difficult to meet the real-time requirements of online monitoring, but the solution is also very sensitive to measurement noise. This challenge is particularly prominent in monitoring the production process of products such as solid compound seasonings, liquid compound seasonings, semi-solid compound seasonings, and food flavorings (liquids), as these materials often have complex compositions and dynamically changing optical properties.

[0093] Therefore, the present application makes a key innovation in technical thinking: the objective of the inversion calculation module is not to reconstruct the complete concentration distribution image, but to directly calculate the spatial covariance matrix of the concentration distribution of the key component, which can more robustly and efficiently reflect the essence of mixing uniformity. The physical meaning of this matrix is that it quantifies the dispersion degree or spatial correlation of the concentration values of each point in the material. In the initial stage of mixing, when the distribution of the key component in the material is extremely uneven, the spatial covariance of its concentration distribution will be large; as the mixing process proceeds, for example, in the "mixing" step in the powder flavoring / seasoning process, the "emulsification stirring" and "homogenization" steps in the emulsified flavor process, or even the "heating stirring" step in the paste (cream) seasoning process, the components tend to be consistent in space, and the numerical value of this covariance matrix will correspondingly and continuously decrease. Therefore, calculating this matrix is the fundamental approach to quantitatively evaluating the above-mentioned process.

[0094] In order to efficiently and robustly achieve this calculation goal, the inversion calculation module adopts an advanced algorithm based on the Bayesian inference framework. The specific calculation process and technical content are as follows:

[0095] The inversion calculation module first solves the inverse problem based on the response data and the forward physical model. This solving process is carried out under a rigorous Bayesian inference framework, and the steps include:

[0096] First, based on the forward physical model and the statistical characteristics of the measurement noise in the response data, a likelihood function is constructed. Specifically, according to the forward model , and reasonable assumptions are made about the statistical characteristics of the measurement noise (e.g., assuming that it follows a zero-mean Gaussian distribution, and its covariance matrix is ), a likelihood function can be constructed. This function describes the probability of obtaining the current response data given an internal concentration distribution .

[0097] Second, a prior probability distribution is set for the concentration distribution of the key component. This step aims to incorporate prior information or physical constraints about the state of the material known before measurement into the calculation to improve the stability and accuracy of the inversion. For example, based on the physical properties of the target product, it can be assumed that its concentration distribution has a certain smoothness in space. This prior knowledge is expressed as a prior probability distribution , and its characteristics are described by a prior covariance matrix .

[0098] Then, under the Bayesian inference framework, the likelihood function and the prior probability distribution are combined to obtain the spatial covariance matrix.

[0099] ;

[0100] wherein, is the posterior probability distribution, which combines the measurement data and the prior information, represents the updated, more accurate knowledge of the internal true concentration distribution after obtaining the measurement data . is the aforementioned likelihood function; is the aforementioned prior probability distribution; denotes the probability is proportional to , indicating that they are related but not identical, and contains a constant factor for normalization.

[0101] Finally, and also the core calculation step of this module, is to calculate the covariance of the posterior probability distribution based on the determined posterior probability distribution, so as to obtain the spatial covariance matrix. As before, the present application ingeniously avoids solving the mean value (i.e. the complete concentration image ) of the posterior probability distribution , but directly calculates its covariance. This calculation is efficiently performed by solving the following posterior covariance matrix formula:

[0102] ;

[0103] wherein, is the estimate of the spatial covariance matrix; is the sensitivity matrix corresponding to the forward physical model; is the transpose of the sensitivity matrix; is the covariance matrix of the measurement noise; is the prior covariance matrix of the key component concentration distribution. The calculation process of this formula only involves matrix multiplication, addition and inversion operations, and the calculation complexity is controllable, which is very suitable for embedding into online monitoring systems that require fast response.

[0104] Through the implementation of this set of advanced algorithms based on the Bayesian framework, the inversion calculation module can stably and efficiently extract the core statistical index (spatial covariance matrix) directly related to the mixing uniformity from the external measurement data.

[0105] The index generation and display module is used to receive the calculated spatial covariance matrix, and generate and output at least one quantitative index for representing the mixing uniformity of the material in the stirring tank based on the spatial covariance matrix.

[0106] In this embodiment, the index generation and display module serves as the last link of the information transmission chain, and its core task is to convert the spatial covariance matrix output by the aforementioned inversion calculation module, which is mathematically rigorous but relatively abstract in physics, into one or more intuitive, quantifiable, and directly guiding production operation mixed uniformity indicators.

[0107] The fundamental purpose of the index generation and display module is to serve as a bridge between complex background calculations and front-end production practices. For solid, liquid, and semi-solid composite seasoning and food essence (liquid) production line operators, a high-dimensional covariance matrix is difficult to understand and use for process judgment directly. Therefore, it is necessary to "reduce" and "concentrate" this matrix, which contains rich material internal state information, into one or several key numerical scalars, thereby making objective and standardized process control possible.

[0108] Further, to achieve this function, in a preferred embodiment, the specific composition and workflow of the index generation and display module are as follows:

[0109] The index generation and display module is first configured to receive the spatial covariance matrix calculated by the inversion calculation module in real time This matrix, as the only input of this module, contains second-order statistical information about the concentration distribution of key components at all discrete voxel positions inside the mixing tank.

[0110] After receiving the matrix, the core function of the index generation and display module is to calculate a numerical scalar based on the received spatial covariance matrix and use it as at least one quantitative indicator for representing the mixing uniformity of the material in the mixing tank. Mapping a matrix to a scalar can significantly simplify the difficulty of information interpretation, so that the mixing state can be represented by a single time-varying numerical value.

[0111] In a specific embodiment, the quantitative indicator can be selected from the following group, or used in combination, to comprehensively evaluate the mixing state:

[0112] A preferred indicator is the trace of the spatial covariance matrix. The trace is calculated as the sum of all elements on the main diagonal of the matrix. In a physical sense, since the main diagonal elements of the covariance matrix represent the variance of the component concentration at each spatial position, the trace of the matrix is proportional to the total variance of the material concentration inside the mixing tank. In the initial stage of mixing, the material distribution is extremely uneven, and the total variance (i.e., the trace) is high; as the mixing process proceeds, for example, in the "mixing" step in the powder essence or seasoning process, the material tends to be uniform, the concentration variance at each point decreases, and the trace of the matrix also monotonically decreases.

[0113] Another optional indicator is the determinant of the spatial covariance matrix. The determinant is known in multivariate statistics as the "generalized variance", which takes into account not only the variance of each point, but also the covariance (i.e. correlation) between different spatial positions. A determinant value close to zero indicates that the distribution of the material concentration in the multi-dimensional space is very concentrated, i.e. highly homogeneous.

[0114] In addition, the matrix norm of the spatial covariance matrix, such as the Frobenius norm, can also be used as a quantitative indicator. The matrix norm can be understood as the "size" or "length" of the matrix. It also integrates the information of all elements (including variance and covariance) in the matrix. The more heterogeneous the mixture, the larger the absolute value of the elements in the covariance matrix, and the larger the norm.

[0115] Finally, the indicator generation and display module also includes the function of displaying the generated quantitative indicators. Preferably, this module is connected to a human-machine interface (HMI), such as an industrial control screen or a computer display. On the display interface, the value of the above calculated quantitative indicators (e.g. the trace of the matrix) can be plotted in real time over time, forming a mixing progress curve.

[0116] By observing this curve, the operator can intuitively judge the mixing state. For example, in the "emulsification stirring" and "homogenization" steps of the emulsified flavor process, or in the "heating and stirring" step of the paste seasoning process, when the curve gradually changes from the rapid decline stage to flat, and the slope approaches zero, it can be objectively determined that the mixing process has reached the required homogeneous state, and thus an instruction can be issued to end the stirring or proceed to the next process. This method provides a clear and reliable basis for determining the optimal process endpoint in the production process of various compound seasoning products, which helps to ensure the stability of the quality of batches of products.

[0117] The present application also provides a stirring tank mixing uniformity test method, comprising the following steps:

[0118] S1, projecting a preset spatio-temporal light field to the material in the stirring tank;

[0119] S2, capturing the response light generated by the material in the stirring tank to the spatio-temporal light field, and generating response data;

[0120] S3, storing a preset forward physical model, the forward physical model being used to characterize the relationship between the response data and the internal key component concentration distribution of the material in the stirring tank;

[0121] S4, based on the response data and the forward physical model, calculating the spatial covariance matrix of the key component concentration distribution by solving the inverse problem.

[0122] S5, receiving the calculated spatial covariance matrix, and generating and outputting at least one quantitative index for representing the uniformity of the mixing of the material in the stirring tank based on the spatial covariance matrix.

[0123] The method of the embodiment can be used to execute the system embodiment described above, and has similar principles and technical effects, which will not be described here again.

[0124] Although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A mixing tank mixing uniformity test system, characterized in that: include: The spatiotemporal light field excitation module is used to project a preset spatiotemporal light field onto the material in the mixing tank; a high-dimensional light field response capture module configured to capture the response light generated by the material in the mixing tank to the spatiotemporal light field and generate response data; A model storage module, configured to store a preset forward physical model, wherein the forward physical model is configured to characterize the relationship between the response data and the concentration distribution of key components within the material in the mixing tank; An inversion calculation module, configured to calculate the spatial covariance matrix of the key component concentration distribution by solving an inverse problem based on the response data and the forward physical model; An index generation and display module, configured to receive the calculated spatial covariance matrix, and generate and output at least one quantitative index for characterizing the mixing uniformity of the materials in the mixing tank based on the spatial covariance matrix; The forward physical model in the model storage module is established based on the radiation transmission theory that describes the propagation process of photons in the material in the stirring tank as a scattering medium. The forward physical model uses a sensitivity matrix F to characterize the relationship between the response data and the concentration distribution of key components in the material in the stirring tank, which is defined as the following mathematical expression: Y=F·X+∈; Wherein, Y is the vector consisting of the response data generated by the high-dimensional light field response capture module; F is the sensitivity matrix corresponding to the forward physical model; X is the vector representing the concentration distribution of the key component; ∈ is the measurement noise vector contained in the response data; The inversion calculation module includes: To solve the inverse problem based on the response data and the forward physical model: constructing a likelihood function based on the forward physical model and measurement noise characteristics in the response data; Setting a prior probability distribution for the key component concentration distribution; Under the Bayesian inference framework, combining the likelihood function and the prior probability distribution to perform calculations to obtain the spatial covariance matrix; The step of calculating, in combination with the likelihood function and the prior probability distribution under the Bayesian inference framework, to obtain the spatial covariance matrix includes: According to the Bayesian inference framework defined by Bayesian theorem, a posterior probability distribution P(X|Y) describing the concentration distribution of the key component is determined, wherein the mathematical relationship of the Bayesian theorem is: P(X|Y)∝P(Y|X)·P(X); Wherein, P(X|Y) is the posterior probability distribution; P(Y|X) is the likelihood function; P(X) is the prior probability distribution; Based on the determined posterior probability distribution, the covariance of the posterior probability distribution is calculated to obtain the spatial covariance matrix, wherein the calculation is performed by solving the posterior covariance matrix formula: Where, is the estimated value of the spatial covariance matrix; J is the sensitivity matrix corresponding to the forward physical model; J T is the transpose of the sensitivity matrix; C ∈ is the covariance matrix of the measurement noise; C prior is the prior covariance matrix of the key component concentration distribution.

2. A stirring tank mixing uniformity testing system according to claim 1, characterized in that: The spatiotemporal light field excitation module includes: at least one light source for generating a modulated probe beam; At least one scanning device is used to control the detection light beam to project toward the material in the mixing tank according to a preset scanning trajectory, thereby forming the preset spatiotemporal light field.

3. A stirring tank mixing uniformity testing system according to claim 1, characterized in that: The high-dimensional light field response capture module includes: receiving the response light generated by the material to the spatiotemporal light field from a plurality of different positions on the surface of the material in the stirring tank; converting the received response light into one or more electrical signals; The one or more electrical signals are processed to generate the response data.

4. A stirring tank mixing uniformity testing system according to claim 1, characterized in that: The indicator generation and display module includes: Receiving the spatial covariance matrix calculated by the inversion calculation module; Based on the received spatial covariance matrix, a numerical scalar is calculated and at least one quantitative index for characterizing the mixing uniformity of the materials in the mixing tank is output, wherein the quantitative index includes at least one of the following group: the trace of the spatial covariance matrix; The determinant of the spatial covariance matrix; The matrix norm of the spatial covariance matrix; The generated quantitative indicators are displayed.

5. A method for testing mixing uniformity in a stirred tank, applied to a system for testing mixing uniformity in a stirred tank as claimed in any one of claims 1 to 4, characterized in that: The following steps are involved: Projecting a preset space-time light field onto the material in the mixing tank; capturing the response light generated by the material in the stirring tank to the spatiotemporal light field and generating response data; Storing a preset forward physical model, wherein the forward physical model is used to characterize the relationship between the response data and the concentration distribution of key components in the material in the mixing tank; Calculating the spatial covariance matrix of the key component concentration distribution by solving an inverse problem based on the response data and the forward physical model; The calculated spatial covariance matrix is ​​received, and at least one quantitative index for characterizing the mixing uniformity of the materials in the mixing tank is generated and output based on the spatial covariance matrix.

Images