AI and the Future of Edible Oil Refining

The oil industry, like many others, is undergoing a transformation driven by the wave of AI technology, aiming to make operations smoother and more productive, from supply chains to the availability of the final product on store shelves.

In a world experiencing significant fluctuations in commodity prices, and with the necessity of adhering to food safety standards and the imperative of sustainability, the oil industry is increasingly adopting Artificial Intelligence (AI) as a core infrastructure, transforming the system from merely "reactive" to a "predictive autonomous" model.

The oil refining process poses a challenge for traditional control systems (PLCs), which struggle to handle non-linear variables in the raw material. This is where AI architectures, especially those utilizing Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTM) models, emerge as qualified tools for processing time-series data to identify patterns that precede equipment failures or deviations in product quality.

The primary objectives of oil refining are to remove impurities such as Free Fatty Acids (FFA), phospholipids, pigments, and oxidation products, while preserving the oil's nutritional, chemical, and physical properties. Traditional refining methods rely on time-consuming laboratory analyses to assess refining efficiency and production quality, which creates delays in feedback loops, often leading to suboptimal quality or excessive chemical consumption.

Examples of AI Applications in Oil Refining Units

• Neutralization and Degumming Unit:AI is integrated with sensors like NIR to measure in-line oil quality (phosphorus, free fatty acids, soap, etc.), and operating condition sensors such as pressure, temperature, and flow rate sensors, etc. This enables optimization decisions aimed at adjusting additives or operating conditions to reduce loss, control quality, and prevent the waste of unnecessary additives that could lead to additional costs.
• Bleaching Unit:The bleaching stage relies on mixing oil with activated bleaching earth to remove pigments and metals. AI-powered bleaching systems use in-line sensors to monitor Lovibond color and metal levels. Through predictive models, these systems can calculate the minimum required earth dosage, reducing consumption compared to traditional operation. AI also automatically manages the filtration cycle using filter differential pressure data to determine the optimal moment to stop the filter for emptying, cleaning, and drying the cake to achieve the highest percentage of recovered oil.
• Deodorization Unit:The deodorization unit is the most energy-intensive part of the oil refining process. Deodorization is carried out by distillation under high vacuum (1.5 to 5 mbar) at temperatures between 240 and 260 degrees Celsius. AI-based process simulation software, such as DWSIM, enables accurate analysis of variables in the stripping tower, ensuring the removal of over 99.99% of free fatty acids while minimizing the formation of trans-fats and 3-MCPD esters. Based on variable analysis, AI algorithms adjust stripping steam consumption, temperatures, vacuum pressure, and oil retention time within the stripping tower, leading to significant energy savings.

AI's Role in Monitoring and Maintaining Equipment Reliability and Production Continuity

AI plays a pivotal role in monitoring equipment and ensuring its reliability and production continuity. This role can be summarized in the following key points:

1. Continuous & Autonomous Monitoring:

• Mobile Robots: The use of quadruped robots (like "Spot"), which have been deployed in Cargill factories to perform thousands of weekly inspections in hazardous or 24/7 operating environments.
• Multi-Sensors: Reliance on thermal imaging and acoustic sensors to detect issues such as overheating bearings, air leaks, or vapor leaks.
• Vision AI: Monitoring high-speed filling lines with microscopic precision to detect packaging defects, leaks, or deviations in filling levels, thereby preventing defective products from reaching the market.

2. Transition to Predictive Maintenance:

• Predicting failures before they occur: Analyzing historical and real-time data (vibration, electrical current, pressure) to identify early signs of failure weeks before the actual breakdown.
• Calculating Remaining Useful Life (RUL): Estimating the remaining lifespan of equipment (such as boilers, pumps, and heat exchangers) before maintenance is required, allowing repairs to be scheduled during planned downtime instead of unexpected outages.
• Preventing "Run-to-Failure": Reducing reliance on corrective maintenance (which occurs after a failure) or rigid periodic maintenance, and instead adopting condition-based maintenance.

3. Process Optimization:

• Digital Twins: Creating simulation models of production lines to test optimal scenarios, thereby increasing Overall Equipment Effectiveness (OEE).

Other Vital Areas:

• Sustainability: AI can also play a significant role in sustainability and reducing the carbon footprint in refining through smart management of wash water for centrifuges and adjusting oxygen-to-fuel ratios to ensure complete combustion.
• Supply Chains: AI can predict the shelf life of inventory based on storage conditions, which aids in efficient smart inventory management and reduces costs, losses, and returns.
• Research and Development: AI can suggest new products and formulations to achieve products that are more stable against oxidation and balanced in terms of fatty acids.

Risks of Total Reliance (Operational and Human Risks)

While AI opens up incredible opportunities for operational improvement, rushing into total reliance on it without proper controls can lead to severe consequences that could cripple the entire production chain.

• Loss of Manual Control: When a factory is entirely run by a single algorithm, any software vulnerability could bring the entire chain to a halt. In traditional systems, human operators can intervene manually, but in fully "self-driving" systems, humans might lose the ability to control.
• The Butterfly Effect: AI makes decisions based on the data it was trained on. If the data contains historical errors, the system will replicate and amplify them, turning troubleshooting into an operational nightmare.
• Human Skill Atrophy: Complete reliance leads to the atrophy of engineers' and operators' skills. If AI performs all tasks for years, who will possess the engineering and scientific acumen to manage a crisis when the system fails? Humans becoming mere screen monitors weakens an organization's ability to innovate and solve non-routine problems.

The Cybersecurity Dilemma (The Biggest Risk)

In modern factories, integrating Industrial Internet of Things (IIoT) with AI expands the scope of risks, such as:

1. Ransomware attacks.
2. Data Poisoning.
3. Digital industrial espionage.

A Final Word: AI is an excellent servant; the secret lies in treating it as a "technical consultant," while always keeping the "kill switch" in the hands of an expert human technician.

‍