Neural Image Foam Detection for Adaptive Bioprocess Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for foam control in bioprocesses, such as microbial fermentation and cell culture, are labor-intensive, prone to errors, and lack flexibility and robustness, often leading to suboptimal control and product quality issues due to manual or fixed-time antifoam agent addition, and existing foam sensors require dedicated integration and calibration.
Innovation Solution
A computer-implemented method using a deep neural network classifier to detect foam on the surface of a liquid medium in a vessel through image classification, which is flexible, accurate, and robust, allowing for foam detection without specialized equipment or fixed light conditions, and can trigger appropriate interventions based on foam levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a dedicated light source and camera system is integrated with the bioreactor for foam detection, then foam detection capability is improved, but device complexity and ease of operation deteriorate due to extensive calibration requirements and integration needs
Solution Approach 1:
The patent uses a standard camera to capture images of the liquid-gas interface, creating a visual copy of the foam situation. This optical copy is then processed by a deep neural network to detect foam, eliminating the need for specialized sensors while maintaining detection capability. The system captures the visual state of the interface and processes it through image analysis rather than requiring direct physical measurement sensors.
Solution Approach 2:
The patent employs a standard camera, which is a universal device already present in many bioreactor systems for other purposes, to perform foam detection. This multi-functional use of existing equipment eliminates the need for dedicated foam detection hardware, reducing system complexity while maintaining detection functionality.
2Measurement precision
If a dedicated light source and camera system is integrated with the bioreactor for foam detection, then foam detection capability is improved, but ease of operation worsens due to extensive and continuous calibration requirements
Solution Approach 1:
The deep neural network classifier is trained on diverse training images that encompass various lighting conditions, angles, and liquid-gas interface locations. This pre-training enables the system to automatically adapt to different conditions without requiring manual calibration, making the system self-sufficient and eliminating the burden of continuous calibration on operators.
Solution Approach 2:
The patent trains the deep neural network on images with varying parameters including different light intensities, color temperatures, viewing angles, and liquid-gas interface positions. This exposure to parameter variations during training enables the system to maintain accuracy across different operating conditions without requiring recalibration when parameters change.
3Extent of automation
If antifoam agents are added on a fixed time schedule to control foam, then foam control is automated, but bioprocess control quality deteriorates due to excessive antifoam agent addition
Solution Approach 1:
The system implements a feedback mechanism where the deep neural network continuously monitors images of the liquid-gas interface and provides real-time information about foam presence and extent. Based on this feedback, the control system can dynamically adjust antifoam agent addition timing and quantity, adding agents only when and where foam is actually detected, thereby optimizing bioprocess control quality while maintaining automation.
Solution Approach 2:
The deep neural network is pre-trained on extensive datasets of foam and non-foam images before deployment. This preliminary training enables the system to accurately recognize foam conditions from the start, ensuring reliable automated control decisions are made based on accurate foam detection rather than fixed schedules.
4Adaptability or versatility
If manual foam control by technician observation is used, then flexibility in foam control decisions is improved, but productivity and reliability worsen due to labor intensity and human error
Solution Approach 1:
The patent replaces the mechanical system of manual visual observation by technicians with an automated optical detection system using a standard camera and deep neural network image analysis. This substitution maintains the flexibility of visual assessment while eliminating labor intensity and human error, improving productivity and reliability.
Solution Approach 2:
The deep neural network automatically performs the foam detection and classification task that previously required technician expertise. The system serves itself by autonomously analyzing images, identifying foam conditions, and providing control recommendations without requiring human intervention, thereby maintaining decision flexibility while dramatically improving productivity.
Data Source
AI summary
A computer implemented method for detecting foam on the surface of a liquid medium contained in a vessel is described. The method including the steps of receiving a sample image of at least a portion of the vessel comprising the liquid-gas interface and classifying the sample image between a first class and at least one second class, associated with different amounts of foam on the surface of the liquid. The classifying is performed by a deep neural network classifier that has been trained using a plurality of training images of at least a portion of a vessel comprising a liquid-gas interface. The plurality of training images may comprise at least some images that differ from each other by one or more of: the location of the liquid-gas interface on the image, the polar and/or azimuthal angle at which the liquid-gas interface is viewed on the image, and the light intensity or colour temperature of the one or more light sources that illuminated the imaged portion of the vessel when the image was acquired. Related methods for controlling a bioprocess, for providing a tool, and related systems and computer software products are also described.