In the realm of clinical nutrition, Food for Special Medicinal Purposes (FSMP) plays a vital role in patient recovery. Integrating Traditional Chinese Medicine (TCM) into these foods—a concept known as the "homology of medicine and food"—offers unique therapeutic benefits. However, manufacturing these complex mixtures presents a significant engineering challenge: ensuring that diverse herbal powders and granules are mixed with perfect uniformity.
Researchers publishing in IEEE Access have addressed this by developing an "Improved Hill Climbing" (IH) genetic algorithm. This approach optimizes the dry mixing process, balancing physical standardization with advanced algorithmic modeling to ensure dosage safety and nutritional consistency.
The Challenge of Dry Mixing
The study focuses on a dry mixing process, which is preferred over wet processing for its cost-effectiveness and ability to preserve heat-sensitive nutrients. The primary obstacle is "stratification." TCM ingredients vary wildly in density, particle size, and mass fraction. Without optimization, these differences cause ingredients to separate during mixing rather than blend.
To achieve uniformity, manufacturers must fine-tune parameters such as mixer speed, duration, and load. Traditional trial-and-error methods are prohibitively expensive and slow, necessitating a more robust mathematical model.
A Two-Pronged Optimization Strategy
The authors propose a hybrid approach that combines physical standardization with a novel optimization algorithm.
1. Physical Pre-processing
Before algorithmic optimization can occur, the physical state of the raw materials must be standardized. The researchers determined that sieving all materials through a 50-mesh screen significantly reduces variability. Furthermore, ingredients with minute mass fractions, such as vitamins, require a "pre-mixing" phase before being integrated into the bulk formula.
2. The Improved Hill Climbing (IH) Algorithm
The core technical contribution is the IH algorithm, which enhances standard Genetic Algorithms (GA) and Hill Climbing (HC) techniques. Key features include:
- Maximum Support Tree: Utilizes Mutual Information theory to map correlations between variables, effectively limiting the search space.
- Adaptive Probabilities: The algorithm dynamically adjusts crossover and mutation rates based on the proximity of a solution to the global optimum.
- Hybrid Operators: By switching between single-point and double-point genetic operations, the algorithm avoids getting trapped in "local optima"—a common failure point for traditional Hill Climbing.
Experimental Validation
The IH algorithm was tested against standard benchmark networks (Asia, Car, and Alarm) and real-world experimental data. The experimental FSMP formula consisted of 13 raw materials, including whey protein, maltodextrin, and herbal extracts like jujube and coix seed.
The model identifies six critical variables for mixing uniformity ($C_v$):
- Premixing time ($T_y$)
- Premixing speed ($V_y$)
- Loading amount ($M$)
- Total mixing time ($T_z$)
- Total mixing speed ($V_z$)
Key Findings and Future Directions
The IH algorithm demonstrated superior accuracy and structural learning compared to standard GA, HC, and Stochastic Hill Climbing (SHC). In benchmark tests, it consistently identified more correct relationships between variables with fewer errors. While the added complexity of genetic operators makes it slightly slower than basic SHC, the gains in accuracy and the ability to find global optima make it far more suitable for industrial applications.
Future work may involve integrating this model with real-time industrial sensors to allow for dynamic adjustments during production, further bridging the gap between traditional medicine and modern industrial engineering.
Summary updated