دانلود کتاب هوش مصنوعی در مهندسی شیمی

The term artificial intelligence (AI) describes the capacity of machines to behave in ways that appear intelligent, responding to novel inputs by making judgments without explicit programming. While conventional computer programs provide results based on explicit instructions, AI systems are made to make predictions using data-driven models. These AI models learn input-output relationships by initially being trained on representative data sets with known output values. Following training, the models can be used to produce new data or predict output values comparable to the training set. AI has the potential to solve many data related issues with intricate input output linkages that are hard or impossible to model procedurally. In the field of chemical engineering, where complex correlations are frequently present in data sets, AI can be applied to a variety of tasks. For instance, theoretical computations or equations based on empirical data may be used to forecast a new compound’s solubility [1]. As an alternative, an AI algorithm that has learned structure-solubility connections from training on many compounds with known solubilities may also be able to predict solubility. Due to the rapid increase in processing power, open source machine learning (ML) frameworks, and chemists’ growing data literacy [2], the application of AI for tasks like property prediction has increased dramatically in recent years. By enabling laboratory automation, predicting the bioactivities of new pharmaceuticals [3], improving reaction conditions, and providing synthetic paths to difficult target molecules, AI implementations have demonstrably reduced design and experimental effort. AI, often termed machine intelligence, has emerged as a crucial field within computer science and automation. It integrates knowledge from various disciplines, including computer science, automation, information engineering, mathematics, psychology, linguistics, and philosophy. AI challenges are categorized into subproblems based on their specific characteristics or functions. In the realm of green manufacturing, the primary issues can be grouped into three main categories: (1) information integration, (2) dynamic risk assessment and decision-making support, and (3) early warning systems. Several techniques are utilized to tackle these challenges and promote green manufacturing, such as knowledge graphs, Bayesian networks, and deep learning (DL), which will be explored in detail later [4]. In the context of process engineering, where AI is being applied , raw materials are transformed into commercial products that are subsequently employed in associated sectors, like manufacturing, where these intermediate items are combined with other components to produce final goods [5]. The following areas of process engineering activities are closely related: process design, process control, process operations, and supporting tools. Process economics might also be included on this list, which would overlap with the engineering tasks. There are numerous subfields and derivative jobs for each of these fields. For example, life cycle management, process evaluation, and risk management are all included in derived tasks. The fundamental idea behind process engineering is like that of chemical engineering, which was developed in the 19th century. Still, it is used for a wide range of industrial processes that change matter and energy, including those involving pharmaceuticals, biotechnological materials, paper, cement, metal, cosmetics, food, chemicals, and more. PSE, or process systems engineering, is also included [6]. The main objectives of the process engineering applications discussed include data rectification, predictive process modeling, process optimization, fault/anomaly detection, process control, and the creation of mechanistic models. Data correction is a preprocessing step frequently employed before AI training. Predictive modeling, the most crucial objective, is also widely utilized in many applications, mostly as an initial step or process intermediate (e.g., as a basis for future process optimization, process control, fault diagnostics, or mechanistic modeling). In the framework of the objective, mechanistic modeling is special because its main objective is to break down educated empirical data-driven models to produce transparent mechanistic or physical white box models and insights [7]. The long history of AI, increasing interest in, and usefulness of ML and DL in the real world have led to the development of a broad variety of AI methodologies. This chapter provides a broad classification of AI, explains the general operation of neural networks, and briefly presents the most significant variations of AI algorithms. Supervised learning is used in AI applications in over 70% of cases. Unsupervised, semisupervised, and reinforcement learning are also used in 10–20% of cases. Supervised learning differs from unsupervised learning primarily in that it provides output data, primarily in the form of labeled data. As a result of outside help, the intended result is known [8]. Similar to classification, clustering divides incoming data into distinct categories. Unlike classification, the algorithm does not know the output variables because the training data are not labeled. The system gains the ability to group input data according to similar characteristics during training and predict related aspects of fresh input data according to their affiliation with a known group. It is possible to specify the total number of these groupings. The user can assess the underlying importance and ramifications of the qualities that have been detected, or the clusters that have been identified can be utilized as features to improve AI techniques. A data set with less multidimensionality can be created by compressing a set of input data with a high degree of multidimensionality from many sources (like sensors) using dimension reduction. Subsequent data analysis can make use of the generated data. Similar to supervised learning, semisupervised learning is an intermediary technique that simultaneously has access to labeled and unlabeled data. The unlabeled data set is typically greater than the annotated data set. Techniques like classification, regression, and prediction allow the AI to conduct its labeling. With reinforcement learning, the algorithm (the agent) seeks to maximize a reward based on the chosen action. Here, the agent adapts its actions for maximum reward by using trial and error and feedback from the environment [9]. This chapter explores the transformative impact of AI on chemical engineering, focusing on its evolution and current advancements. Traditional approaches are being reshaped by the integration of AI technologies, driving innovation and improving efficiency across the field. ML and data analytics have been proven to enhance safety and optimize processes in various use cases. Looking ahead, AI’s role in modernizing chemical processes will be crucial in addressing complex challenges and promoting sustainable practices. Embracing these advancements is essential for the continued success and growth of the chemical engineering industry. As AI continues to evolve, it holds the potential to revolutionize not only how chemical engineers design, optimize, and monitor processes, but also how they contribute to solving global challenges, such as energy efficiency, resource management, and environmental sustainability. The continued development and adoption of AI in this domain promise to unlock new levels of creativity, precision, and productivity, which will be essential for shaping the future of chemical engineering. This chapter presents a comprehensive overview of the latest developments and the potential of AI to revolutionize the field of chemical engineering, supporting innovation and sustainability in the industry.