Authoritative Overview! Unveiling 4 mainstream fermentation control strategies and 3 core challenges, covering all scenarios of food and medicine ..
This review paper published in Chemical Engineering Technology systematically outlines the core challenges and technological frontiers of fermentation process control.
The difficulty in controlling the fermentation process mainly stems from three core challenges
Highly nonlinear and complex: The growth and metabolic behavior of microorganisms is not a uniform linear motion, but a complex process full of "temperament". Even a slight change in temperature or pH can cause them to "switch" metabolic pathways and produce completely different products.
Cognitive 'black box': Our understanding of the complex biochemical reaction mechanisms within microorganisms is still incomplete, just like facing a partially transparent black box, it is difficult to establish accurate mathematical models to predict their behavior at every step.
Measurement tool lag: Lack of reliable, economical, and real-time online monitoring sensors for key indicators (such as specific cell concentration and product concentration), resulting in the control system being unable to obtain timely feedback.
Fermentation mode: batch, feed and continuous production
Before choosing a control strategy, it is necessary to first understand the three basic "combat modes" of fermentation, each with its own advantages and disadvantages and suitable for different scenarios.
Batch fermentation: one-time feeding, one-time harvesting after completion. The advantages are simple operation and low pollution risk; The disadvantage is low product yield and long preparation time between batches.
Supplementary fermentation: During the process, fresh culture medium is continuously added without discharging any products. The advantage is that it can avoid substrate inhibition, increase yield, and control by-products; The disadvantage is that the streaming strategy is complex and may result in genetic instability.
Continuous fermentation: continuous feeding, continuous discharging, maintaining a constant volume inside the reactor. The advantage is high output and no downtime; The disadvantages are complex downstream processing, high pollution risk, and susceptibility to genetic instability.
Faced with challenges, engineers have developed various advanced control strategies that resemble commanders of different styles.
1. Model Predictive Control - MPC, a visionary planner, is an advanced control strategy widely used in industry.
Its core idea is to predict the future and optimize the present.
Working principle: MPC utilizes mathematical models of processes to predict how key parameters will change in the future (in the prediction time domain). Then, it calculates a series of optimal control actions (such as adjusting the feed rate), but only implements the first step. Next, the system updates the prediction based on the latest actual measurement values and recalculates the optimization, scrolling through the process.
Advantages: Able to "foresee" future changes, take action in advance, and directly handle various constraints.
Challenge: Its performance heavily relies on the accuracy of the model and requires a large amount of computation.
The paper mentions that NMPC has demonstrated better performance than traditional control in both penicillin production and glutamic acid fermentation, and can achieve higher final product concentrations.
2. Fuzzy Logic Control - Experienced 'Master'
Fuzzy logic control does not rely on precise mathematical models, but imitates the experience and thinking of human experts.
Working principle: It converts vague language descriptions such as "water temperature is a bit high" and "feed too fast" into rules that can be processed by computers. The system determines how to operate through a series of 'if... then...' rules.
Advantages: Very effective and robust for complex nonlinear systems that are difficult to model accurately.
Challenge: The establishment of a rule base relies on expert knowledge and lacks learning ability.
3. Meta heuristic Algorithm - Nature's "Optimization Master"
This type of algorithm simulates natural phenomena such as genetics and bird foraging to find the optimal solution.
Working principle: Taking genetic algorithm as an example, it encodes possible solutions into "chromosomes", simulates natural evolution processes such as "selection, crossover, and mutation", eliminates bad solutions generation by generation, preserves excellent solutions, and ultimately approaches the global optimum.
Advantages: It does not depend on the specific mathematical properties of the problem, has wide applicability, and can handle complex constraints.
Challenge: May fall into local optima and have high computational costs.
4. Artificial Neural Networks and Reinforcement Learning - AI with "Learning" Ability
Artificial neural network: simulates the neural network of the human brain, trains on a large amount of data, and self learns the complex relationship between input and output. It can be used to establish soft sensors or directly as a controller.
Reinforcement learning: Simulating the "trial and error learning" mechanism. An RL agent learns the optimal strategy based on its action outcomes (rewards or punishments) by interacting with the environment. Just like training a pet, giving rewards for doing things right and punishments for doing things wrong, it can eventually learn to complete complex tasks.
