Intelligent control system of bioreactor

Abstract

The invention provides an intelligent control system for a bioreactor. The control system comprises a data collection module, a time alignment module, a data processing module and a data output module, the data collection module is connected with the image acquisition device and the data acquisition device, and the data acquisition device comprises a plurality of sensors for monitoring the reaction of the culture container; the time alignment module is used for ensuring that the data output to the data processing module are data at the same moment; and the data processing module receives the image data and the sensor data transmitted by the time alignment module, performs feature extraction and splicing, fuses the data transmitted by the processing module by adopting a cross attention mechanism, and transmits a result to the data output module.

Description

Technical Field

[0001] The present invention relates to the field of bioreactors, and particularly to an intelligent control system for a bioreactor based on a neural network. Background Art

[0002] A bioreactor is a device used for carrying out biochemical reactions and is widely applied in the fields of bioengineering, pharmaceuticals, food processing, etc. It promotes microorganisms, cells or enzymes to perform specific biotransformations or produce target products by providing suitable environmental conditions (such as temperature, pH, oxygen, etc.). During the operation of the reactor, it is necessary to dynamically regulate the actions of the bioreactor according to the working parameters so that the reaction always maintains the optimal state.

[0003] The control of existing bioreactors mainly relies on manual control. For example, by taking pictures of the bioreactor and performing manual counting, online real-time control cannot be achieved. Secondly, there are many control parameters for bioreactors, such as temperature, dissolved oxygen, pH, etc. For a stirred tank reactor, it also includes the rotation speed of the stirring paddle, etc. Manual experience relies on single-sensor data (such as temperature, dissolved oxygen, volume of medium replacement / supplementation) for feedback control. The control method based on fixed rules or simple models is difficult to capture complex non-linear relationships, with a high control cost and unable to meet the requirements of real-time online control.

[0004] In addition, the real-time precise control of the rotation speed of the stirring paddle and the perfusion speed (the discharge speed and discharge volume of the culture medium and waste liquid) is still a difficult point in current bioreactors. If the rotation speed of the stirring paddle is too low, the culture medium in the tank will not be mixed sufficiently, and cells cannot expand in the best state. If the rotation speed of the stirring paddle is too high, the shear force of the paddle blade will be too large, which will damage the cells and reduce the cell viability. Existing automated bioreactors will set a fixed operation program, for example: controlling the rotation speed of the stirring paddle to reach a certain value at a certain moment, or controlling the perfusion speed to reach a certain value at a certain moment. The disadvantage of this method is that it cannot perform real-time online regulation according to the state of cell aggregates. Summary of the Invention

[0005] To solve the above problems, the present invention provides an intelligent control system for a bioreactor. The bioreactor includes a culture container, an image acquisition device, and a data acquisition device. The culture container is made of a transparent material or is provided with an observation window. The image acquisition device is used to acquire images inside the culture container. The control system includes a data collection module, a time alignment module, a data processing module, and a data output module; the data collection module is connected to an image acquisition device and a data acquisition device, and the data acquisition device includes a plurality of sensors for monitoring the reaction of the culture container; the time alignment module is used to ensure that the data output to the data processing module is data at the same moment; the data processing module receives the image data and sensor data transmitted by the time alignment module, performs feature extraction and splicing, adopts a cross-attention mechanism, fuses the data transmitted by the processing module, and transmits the result to the data output module.

[0006] Further, the data acquisition device includes a dissolved oxygen sensor, a pH sensor, and a temperature sensor.

[0007] Further, the data processing module includes an image data feature extraction module, a sensor data preprocessing module, and a multi-modal fusion module; The image data feature extraction module includes an image feature extraction layer for extracting features from images; the sensor data preprocessing module vectorizes and extracts features from the data collected by the sensors respectively, and performs sensor data feature fusion; the multi-modal fusion module includes an image self-attention layer, a sensor data self-attention layer, an image-to-sensor cross-attention layer, a sensor-to-image cross-attention layer, and a decoding layer; The data output module includes an output layer, and the output layer receives the data from the above decoding layer and outputs reactor control parameters.

[0008] Further, the reactor state control parameters include the diameter of each biological mass, the estimated remaining culture time, the warning of bacterial contamination risk, the speed of the stirring motor , the speed of the liquid changing / supplying pump , , the unit of is revolutions per minute, -1 .

[0009] Further, the speed of the stirring motor is obtained by the following formula: , The speed of the liquid changing pump is obtained by the following formula:

[0010] where a1, a2, a3, b1, b2, b3, b, and d are all control parameters, which are constant values determined by multiple experimental fittings, is the intermediate value; C avg is the average size of the mass,C rsd is the uniformity of the aggregate size C ratio is the deposition ratio of the aggregates

[0011] wherein M is the total number of biological aggregates C i represents the diameter of the i th biological aggregate among all biological aggregates C j represents taking the first M min biological aggregates arranged in ascending order of diameter and taking the diameter of the j th biological aggregate C k represents taking the first M max biological aggregates arranged in descending order of diameter and taking the diameter of the k th biological aggregate; , , M up and M down are respectively the number of biological aggregates in the upper half and the lower half of the culture vessel.

[0012] Furthermore, the control system adjusts the operating state of the bioreactor according to the reactor control parameters, or a person manually adjusts the operating state of the bioreactor with reference to the reactor control parameters.

[0013] The present invention adopts a bioreactor control system based on a neural network. A series of feature extraction layers and self-attention layers are set in the data processing module. Multimodal data analysis is adopted. Through the fusion of image features (biological aggregate morphology) and sensor data (dissolved oxygen, pH, temperature), the comprehensiveness of state perception is improved; an attention cross mechanism is adopted to automatically discover the implicit association between the image and sensor data (such as the influence of the change in stirring speed on the dispersion degree of aggregates), enhance the adaptability of control decisions, and realize the accurate estimation of parameters such as the number of organisms and the remaining culture time through the decoding of the fused features by the neural network. In addition, according to the estimation of the diameter of the biological aggregates, the speed control parameters of the stirring motor and the liquid exchange pump are obtained, and the rotation speed of the stirring paddle and the perfusion speed are controlled in real time. The biological survival rate and amplification effect are greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a schematic diagram of the bioreactor control system architecture; Figure 2 It is a schematic diagram of the system architecture of the data output module. Specific implementation manners

[0015] A bioreactor generally includes a culture container for containing a culture solution and cells. In order to obtain the reaction conditions inside the culture container, the culture container is made of a transparent material, such as glass or transparent plastic. An image acquisition device (such as a camera) is placed beside the culture container to take pictures of the reaction conditions of the culture container, obtain images of biological aggregates (cell aggregates or microbial aggregates) and analyze them. At the same time, the reactor also includes a data acquisition device (usually various sensors) for obtaining parameters of the reactor, such as temperature, oxygen concentration, etc. According to the current conditions of the biological aggregates and the corresponding parameters, reaction control parameters (such as intake air volume, temperature, stirring speed, etc.) are controlled.

[0016] See Figure 1 In addition, the present invention also provides a control system for a bioreactor, and the control system includes a data collection module, a time alignment module, a data processing module, and a data output module.

[0017] The data collection module is connected to the image acquisition device and the data acquisition device. The data acquisition device includes a plurality of sensors for monitoring the reaction of the culture container, including a dissolved oxygen sensor, a pH sensor, and a temperature sensor, so as to obtain image data, dissolved oxygen data, pH data, and temperature data.

[0018] The time alignment module is used to ensure that the data output to the data processing module is data at the same moment. Because in the data acquisition process, the acquisition frequencies of different acquisition devices are different. For example, the acquisition frequency of the image acquisition device is generally 15 - 30 Hz, while the acquisition frequencies of temperature, pH, and dissolved oxygen are about 50 - 100 Hz. In order to ensure data consistency, it is necessary to ensure that different types of data are acquired at the same moment.

[0019] The data processing module includes an image data feature extraction module, a sensor data preprocessing module, and a multi-modal fusion module.

[0020] The image data feature extraction module uses an image feature extraction layer to extract features from the image.

[0021] The sensor data preprocessing module vectorizes and extracts features from dissolved oxygen data, pH data, and temperature data respectively, and performs sensor data feature fusion. For dissolved oxygen data, pH data, and temperature data, vectorization is required before inputting into the neural network to unify the input dimensions. Specifically: Rescale the dissolved oxygen data, pH data, and temperature data to [0, 100], discretize each data item, and convert it into a vector with a dimension of 100 using the one-hot encoding method. In data feature extraction, input the above vectors into the corresponding feature extraction layers (the input dimensions of the feature extraction layers are all 100), and the output of each feature extraction layer is a vector with a dimension of 100. The sensor data feature fusion layer concatenates these three types of data (dissolved oxygen data, pH data, and temperature data) to form a [3, 100] matrix.

[0022] The multi-modal fusion module adopts a cross-attention mechanism to integrate images, dissolved oxygen data, pH data, and temperature data, and automatically discovers the correlations between the data. The multi-modal fusion module includes an image self-attention layer, a sensor data self-attention layer, an image-to-sensor cross-attention layer, a sensor-to-image cross-attention layer, and a decoding layer. The image self-attention layer receives the data from the image feature extraction module, and the sensor data self-attention layer receives the data from the sensor data preprocessing module. After processing respectively, they are pushed to the image-to-sensor cross-attention layer, and after being processed by the image-to-sensor cross-attention layer, they are pushed to the sensor-to-image cross-attention layer, and then the processed data is sent to the decoding layer.

[0023] See Figure 2 , the data output module includes an output layer, and the output layer receives the data from the above decoding layer and outputs reactor control parameters such as the diameter of each biological mass, the remaining culture time, the warning of bacterial contamination risk, the speed of the stirring motor, and the speed of the liquid changing pump.

[0024] Among them, the remaining culture time and the warning of bacterial contamination risk are directly output to the terminal display, and the speed of the stirring motor is further obtained using the diameter of each biological mass and the speed of the liquid changing pump , and the specific method is as follows: Calculate the uniformity of the mass size Crsd, the average mass size Cavg, and the deposition ratio of the mass Cratio.

[0025] Specifically:

[0026] Where M is the total number of biological masses, C i represents the i th biological mass diameter among all biological masses,C j represents the diameter of the M min -th bioblock when arranged in ascending order of diameter, j and C k represents the diameter of the M max -th bioblock when arranged in descending order of diameter; k

[0027]

[0028] M up and M down are the number of bioblocks in the upper half and the lower half of the culture vessel respectively.

[0029] Furthermore, based on the above values, further determine the speed of the stirring motor of the culture vessel and the speed of the liquid changing pump , where the speed of the liquid changing pump determines the perfusion speed (or liquid changing speed).

[0030]

[0031] In the above formula, a1, a2, a3, b1, b2, b3, b, and d are all control parameters, which are constant values obtained by fitting multiple sets of experimental data, and is the intermediate value.

[0032] The unit of is revolutions per minute, a positive value represents clockwise rotation, and a negative value represents counterclockwise rotation; The unit of is days -1 , representing the culture system multiple. For example, if 2L needs to be cultured and 1L needs to be produced per day, .

[0033] The present invention adopts a bioreactor control system based on a neural network. A series of feature extraction layers and self-attention layers are set in the data processing module. Multimodal data analysis is used. Through the fusion of image features (bioblock morphology) and sensor data (dissolved oxygen, pH, temperature), the comprehensiveness of state perception is improved; an attention cross mechanism is adopted to automatically discover the implicit correlation between images and sensor data (such as the impact of stirring speed changes on the dispersion degree of bioblocks), enhance the adaptability of control decisions, and through the decoding of the neural network for the fused features, accurate estimation of parameters such as the number of organisms and the remaining culture time is achieved.

[0034] In addition, based on the estimation of the diameter of the bio - mass, the speed control parameters of the stirring motor and the liquid - changing pump are obtained to control the rotation speed of the stirring paddle and the perfusion speed in real - time, greatly improving the biological viability and the amplification effect.

[0035] The above - mentioned embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent control system for a bioreactor, the bioreactor including a culture container, an image acquisition device, and a data acquisition device, the culture container being made of a transparent material or provided with an observation window, the image acquisition device being used to acquire images inside the culture container, characterized in that, the control system includes a data collection module, a time alignment module, a data processing module, and a data output module; the data collection module is connected to the image acquisition device and the data acquisition device, and the data acquisition device includes a plurality of sensors for monitoring the reaction of the culture container; the time alignment module is used to ensure that the data output to the data processing module is data at the same moment; the data processing module receives the image data and sensor data transmitted by the time alignment module, performs feature extraction and splicing, adopts a cross-attention mechanism, fuses the data transmitted by the processing module, and transmits the result to the data output module.

2. The control system according to claim 1, wherein The data acquisition device includes a dissolved oxygen sensor, a pH sensor, and a temperature sensor.

3. The control system according to any one of claims 1-2, characterized in that, The data processing module includes an image data feature extraction module, a sensor data preprocessing module, and a multi-modal fusion module; The image data feature extraction module includes an image feature extraction layer for extracting features from images; the sensor data preprocessing module respectively vectorizes and extracts features from the data collected by the sensors, and performs sensor data feature fusion; the multi-modal fusion module includes an image self-attention layer, a sensor data self-attention layer, an image-to-sensor cross-attention layer, a sensor-to-image cross-attention layer, and a decoding layer; The data output module includes an output layer, and the output layer receives the data of the above decoding layer and outputs the reactor control parameters.

4. The control system according to claim 3, wherein The reactor control parameters include the diameter of each biological mass, the estimated remaining culture time, the early warning of contamination risk, and the speed of the stirring motor , the speed of the medium change / refilling pump , The unit of is revolutions per minute, and the unit of -1 is days 5. The control system according to claim 4, characterized in that: The stirring motor speed is obtained by the following formula: ， The liquid exchange pump speed is obtained from the following formula: ， Among them, a1, a2, a3, b1, b2, b3, b, and d are all control parameters, which are constant values, and are intermediate values; C avg is the average size of the agglomerates, C rsd is the uniformity of the agglomerate size, C ratio is the deposition ratio of the agglomerates, ； Among them M is the total number of biological clumps, C i represents the diameter of the i th biological clump among all biological clumps, C j represents taking the M min th biological clump among the first j biological clumps arranged in ascending order of diameter, C k represents taking the M max th biological clump among the first k biological clumps arranged in descending order of diameter; ， ， M up and M down are the number of biological aggregates in the upper half and the lower half of the culture vessel, respectively.

6. The control system according to any one of claims 4-5, characterized in that, the control system adjusts the operating state of the bioreactor according to the reactor control parameters, or the personnel manually adjust the operating state of the bioreactor based on the reactor control parameters as a reference.

Images