Artificial Intelligence for quality control: fairy tale or real added value for organizations?
Data on the market trends associated with artificial intelligence (AI), as previously indicated in one of our earlier articles, reveal that, in both the agricultural and manufacturing sectors, the most advanced frontier (and the one with the greatest potential for generating differential value for companies) is quality control.
The quality of products has been the subject of scrutiny for several years, with instances of food poisoning and high-profile incidents (such as the recall of the blue mozzarella) prompting concern among the general public. Thanks to these notable events, significant progress has been made in regulatory oversight to guarantee the safety of products for final consumers. However, despite these developments, the current regulatory tools remain insufficient to fully prevent the distribution of unsafe products.
Until recently, the only method available to a company (manufacturing or agricultural) to verify the quality of its end products was laboratory spot checks. While these tests are statistically valid, they do not provide a guarantee or certification of the quality of any specific product placed on the market.
In this area, technological evolution offers new opportunities, thanks to the increasingly effective introduction of the Internet of Things (IoT), which enables the use of sensors capable of recording a wide range of signals directly in production, without affecting the product in any way. These signals can then be retained in the production process until commercialization. Additionally, data science systems are capable of transforming these signals into key insights that are understandable and easy to act upon (through targeted interventions or automated corrective measures in process lines).
However, what are the tools currently employed by leading companies? This article will examine a selection of these tools and their potential applications to specific cases.
Spectroscopy analysis in manufacturing
NIR spectroscopy has been a viable alternative to classical analytical methods for a considerable period of time. Near-infrared (NIR) analysis employs near-infrared radiation to transform an electrical signal into an analytical result. The sample is exposed to radiation that is harmless to living organisms. This radiation vibrates the molecules in the sample, which react differently depending on the concentration of various analytes. The resulting signals are mathematically processed, and the concentrations of the various analytes are determined. There is no need for toxic consumables or reagents that would then have to be disposed of, nor for sample preparation, since the product to be analyzed is scanned in-line, providing instantaneous information without interrupting the production process.
These data are automatically integrated into a process optimization platform, such as the SpectralizeR tool, where they become part of a machine learning model (integrated with process indicators) that can retroactively assign each product characteristic, revealed by spectroscopy, to a specific plant activity.
Such integration is not limited to any specific production environment; it can occur in any setting in which spectroscopy (and not just NIR spectroscopy, but anyone based on whatever harmless radiation to the product) can be employed as a quality control, identification and prevention mechanism for manufacturing drifts or errors.
Subsequently, spectroscopy can be integrated with artificial intelligence, which processes the acquired data in conjunction with other process information and responds instantaneously, thereby optimizing the process.
Currently, SpectralizeR (designed to work with any type of probe and radiation) is already in use at a company operating in the agronomy sector. The project is designed to monitor the moisture content of different types of products in real-time at multiple stages of the production process and before the product is bagged, thus ensuring the quality of each product brought to market.
Nevertheless, spectroscopy is not exclusively applicable to organic products. With different types of rays and the development of bespoke AI models, the same analytical techniques can be applied to any type of product. At Kode, for example, XRF (X-ray fluorescence spectroscopy) has already been tested in the field of metallurgy, with promising results.
Micro-Sensors in agrifood for the quality of wine and oil.
A novel system for the rapid assessment of polyphenols (a measure of fruit ripeness) has been developed as a result of a research project in which the Kode team collaborated with the Department of Agricultural, Food and Agri-Environmental Sciences of the University of Pisa, the University of Tuscia, and CNR-Nano Pisa.
The tool developed is an example of how quality controls can be initiated at the earliest stages of the production process, namely from raw materials. This entails direct scanning of the grapes or olive juice at the vineyard or olive grove.
This complex system is based primarily on the innovative sensors studied (based on shear-horizontal SAW wave generation), which can monitor molecular adhesion in a liquid environment in real time. This feature allows the instrument (unlike previous prototypes) to be integrated directly into winemaking or milling equipment.
The data acquired by the sensors is processed by a comprehensive cloud computing system, which enables farmers, winemakers, and operators at various stages of production to visualize the data acquired from vineyard and cellar analyses in real-time, regardless of location or connection tool. In addition to detailed data and its temporal evolution, the tool provides a clear indication of the degree of grape ripening, maceration/fermentation and wine evolution. This same technique, when applied in the field of oil production, has proven useful in monitoring the maturation of olive juice as well as oil during the extraction process.
Each agricultural product possesses distinctive chemical and physical attributes, which must be considered on a case-by-case basis. This project serves as an illustrative example of the advancement of sensor technology and the potential for devising solutions that leverage the latest technological developments, even in areas that may be perceived as low-tech.
Computer Vision for product monitoring.
Just as chemical analysis of materials is not the only way to test the quality of products, computer vision is not just for futuristic autonomous driving systems.
Indeed, while the concept of quality control is often associated with food, it is also crucial in products whose quality is not primarily determined by their chemical and physical composition as by their manufacture, shape or other aspects that can only be verified through careful visual inspection
This encompasses a wide range of products and semi-finished products in the clothing industry, from tanning and textile production to the final product, which is then sold in stores.
However, human visual inspection, naturally imperfect, can be supported and optimized through computer vision. Indeed, through very high-resolution micro-cameras, which can be installed directly on board production or packaging lines, detailed images can be acquired that enable the training of AI models, capable of recognizing specific details of a product through Image Recognition Systems. At Kode, we have applied these techniques in a variety of settings, including cleaning services or the identification of lichen species (recognising thus the different species), as well as in the monitoring of inventory in a warehouse setting.
By integrating computer vision with process control software, quality control is able to flag the defective product or semi-finished product (to be corrected or removed from the market, depending on the production stage at which the control is applied), turning it into a true real-time intervention system.
Conclusions
Many of the most striking developments in the field of quality control (not only in agribusiness) are directed toward the creation of Internet of Things (IoT) systems, which leverage increasingly precise sensors to enable a reduction in the cost of ex-post quality control and, on the other hand, faster data acquisition and its transformation into key operational indications to reduce waste and optimize work.
Data Science is the main tool that has been intervening for years in many industries to structure process or production data acquisition systems to transform all these new data collected by sensors with the available masses of data (often unstructured) into indicators that allow companies to choose actions to be taken on a real basis.
This is how the most forward-looking companies in sectors such as manufacturing, but not only, manage to optimize their operations and continue to grow despite (or despite) fluctuations in the environment. This seems to be the path, with all its peculiarities, even for those sectors that live or have lived without any technological intervention, such as agriculture, whose late start in digitalization does not seem to be a disadvantage today.
Today’s industries that want to challenge themselves with new technologies can take advantage of the experience and research of the past decade, which has finally made stable, high-performance tools and solutions available.
