A research team from the University of Cadiz has applied two different artificial intelligence tools to compare the precision in the quality control process of petrol and classify it according to its capacity for self-combustion. Specifically, an ‘electronic nose’, designed by the same team of experts to detect the remains of flammable liquids in a fire, and infrared have been used to discriminate between the two most commercially available types of petrol.
This advance will allow the industry to have up-to-date, accurate, and real-time information on the composition of this hydrocarbon. The work has been carried out with funds from the Regional Ministry of Economic Transformation, Industry, Knowledge and Universities, the University of Cadiz’s resources, and those of the Wine and Agri-Food Research Institute (IVAGRO), as well as with support from the ERDF.
The researchers have worked with two analytical methods based on the identification of patterns in massive data and the elaboration of predictions, known as ‘machine learning. They have applied them separately and jointly. Thus, they have analysed the usefulness of both methods in two different processes, one based on algorithms – instructions that process data – and the other obtaining chemical measurements of petrol. They have also evaluated the effectiveness of their combination to discriminate and classify hydrocarbon samples.
While the ‘electronic nose’ provides data on the volatile profile of the samples, the spectroscopic techniques focus on analysing non-volatile compounds. The combination of information from both methodologies is used to generate predictive models to discriminate and classify gasoline samples according to their octane rating. “This combination represents a real alternative for automating the quality control process of this petroleum derivative, which currently depends on the experience of the analyst who performs this work,” explained University of Cadiz researcher Marta Barea, who is responsible for this work.
By providing concrete information, the applications of this new methodology in the petrochemical industry contribute greatly to optimising quality processes, as well as in other areas. “With this data, refineries will have fast, on-the-spot quality management systems with a very precise level of detail,” he added.
It is also very useful in the field of forensic chemistry if, for example, a fire breaks out and it is necessary to determine which flammable liquid caused it and from there to follow clues to locate its origin.
The identification models generated in this study can be used to create web applications for computers, tablets, and mobiles and facilitate the automation of the quality processes of this petroleum derivative.
To obtain these results, the research team used data from a total of 50 samples of 95 and 98 octane petrol, which they analysed using these two techniques. First, the model was trained by introducing it to the full set of information it can access. To test whether it was also able to interpret new samples, other data unknown to the model were included.
In conclusion, the experts have obtained good performances in both algorithms, allowing them to classify and determine the samples correctly. However, they noted that the ‘electronic nose’ provides more accurate information on the octane rating of petrol due to the classification of volatile compounds.