Intelligent Process Automation for Industry 4.0

In recent years, there has been a growing interest in Robotic Process Automation (RPA), technologies useful in automating business processes. The interest arises from the fact that they permit to reduce costs and relieve more and more employees from performing repetitive operations of low intellectual content, thus enabling the reallocation of these resources to more valuable processes/tasks. RPA is a technology that assists in the automation of administrative tasks by means of hardware-software bots. Such bots utilise the user interface to capture data by manipulating it as a human would do. However, the time has shown that RPA technologies have limitations, as they only allow the automation of repetitive and structured processes, in fact as soon as these deviate from the standard case, robots are no longer able to handle them. These issues have determined the evolution of this technology through the incorporation of artificial intelligence models and machine learning. This has given rise to an automation 2.0 called Intelligent Process Automation (IPA), capable of increasing the panorama of applications that can be automated, thanks to the possibility of dynamic adaptation that the robot now offers. Poste Italiane has also embarked on a process of automating its services by adopting, among others, RPA in the back office sector for processing and treatment of documents sent by customers to Poste Italiane's institutional mailboxes. RPA alone, however, has a number of limitations in handling unstructured, handwritten, and scanned documentation. This results in automation bringing no real benefit, as documents of this type make up the majority of documentation sent by customers. This thesis project aims to the implementation of an IPA solution capable of overcoming these Document Understanding issues. This was done by introducing models based on machine learning with the goal of maximising the number of documents successfully processed by the robot and thus the benefits of automation. Specifically, in this thesis, the IPA approach allows to extract particular fields (e.g. postcode) which could not be extracted with a traditional RPA approach, from a dataset of Poste Italiane documents. This process was implemented by comparing various available techniques within the UiPath suite and by selecting the best possible solution for the automatic extraction of the fields under analysis. This solution was subsequently retrained on a dataset containing additional documents representing borderline cases because they were, for example, rotated or blurred.