The industrial sector, like other areas, is currently undergoing a phase of digital transformation. This means that manufacturing companies are involved in various digitalization activities [1]. Within this context, industrial data and the way it is processed, visualized, and utilized play an essential role.
Content
Relying solely on technology does not generate business value if industry problems are not thoroughly examined. There are many ways in which industrial AI can contribute to the digital transformation of manufacturing. Some of the most appealing areas where it can be used are: process applications to improve productivity (i.e., intelligent production), product and service applications, insight applications for knowledge discovery (i.e., root cause analysis, decision-making) [2].
Concrete examples that fit the categories mentioned above are two of our ongoing EU projects: “Customizable AI-based in-line process monitoring platform for achieving zero-defect manufacturing in the PV industry” (Platform-Zero) and “Data and Metadata for advanced Digitalization of Manufacturing Industrial Lines” (metaFacturing). The Platform-Zero project aims to improve the production quality of photovoltaic systems while reducing manufacturing costs through zero-defect manufacturing. This is achieved through the application of non-destructive testing methods and technologies to detect, correct, and prevent critical production errors early. The data is evaluated in real-time to optimize the production process and improve product quality. The metaFacturing project focuses on creating a digitized tool chain for the production of metal parts (casting and welding). Trusted AI and hybrid methods are analyzed and implemented to gain process insights, improve production process efficiency (e.g., optimizing process parameters), and enhance product quality (e.g., reducing defects).
To gain a comprehensive understanding of the topic, several key aspects are considered below, categorized into four groups. First, the problems and needs of manufacturing companies are discussed. Additionally, we delve into the added value that industrial AI can offer as a solution to their problems, as well as the challenges that arise when applying industrial AI (see Figure 1).
Figure 1: Core points of the application perspectives of Industrial AI for manufacturing companies
Below we consider some of the key pain points that manufacturers face when it comes to successfully introducing and using industrial AI:
Modern manufacturing today generates a multitude of data through the use of technical systems (sensors, cameras). This results in a mix of structured and unstructured data that is often so complex and extensive that it is difficult to identify clear patterns and insights. Additionally, it is hard to determine which data is relevant for further analysis and should be stored. A concrete example is the production of semiconductors. Enormous amounts of sensor data, process data, and quality data are generated. This data must be analyzed to identify deviations or errors in the production processes. Due to the multitude of data sources and formats, it will be a challenge to extract and interpret the relevant information.
Industrial processes are often characterized by numerous process parameters and interactions that make it challenging to recognize hidden relationships between the data. Take the example of casting parts: In this context, the interrelationships of process parameters – such as melt temperature, mold temperature, first phase speed, and second phase speed – are complex, non-linear, and contradictory [3].
Decision-making in manufacturing requires consideration of numerous factors and constraints. As we saw in the previous section, determining the optimal tolerance window for parameters is not a simple task. Staying in the same context – the casting of metal parts – we find that this decision affects process efficiency (i.e., metal parts are automatically qualified as scrap by the casting machine if the measurements fall outside the tolerance window).
Customer requirements are constantly changing, and manufacturers must be agile to adapt products and meet market demands. The choice of materials has a significant impact on product quality. The transition from traditional steel to advanced lightweight materials such as carbon fiber composites for body panels, for example, brings complex production processes due to the unique properties of carbon.
By identifying the root cause of a problem, manufacturers can implement targeted solutions to prevent the problem from recurring. This requires delving deep into the data to discover correlations and thus the underlying factors likely responsible for certain problems or anomalies. For example, in an electronics manufacturing scenario where a particular batch of products repeatedly fails quality tests, root cause analysis might reveal that a specific machine component is likely not calibrated correctly.
Advanced analysis of production data can detect problems early, before they lead to serious defects or failures. In energy generation, for example, unusual deviations in a generator’s power consumption could indicate a potential problem that needs to be addressed to prevent a breakdown.
Companies are shifting their focus from accumulating mass data to strategic industrial data management. Optimizing data needs and processing is also in line with the goals of the European Commission. Intelligent data selection and preparation reduces the need to collect, store, process, and transfer large volumes of data and/or large AI models, thereby reducing energy consumption [4].
Manufacturers use data analytics to make informed decisions. A chemical plant can gain immediate insights from seamlessly integrated industrial data that spans from the edge to the cloud. This can be achieved by merging various data sources, fostering agile decision-making across the enterprise. In complex decision-making scenarios, this data can also be integrated into optimization models (see https://www.risc-software.at/fachbeitraege-die-besseren-entscheidungen-treffen-dank-prescriptive-analytics/) to support decision-makers with planning problems.
Innovative approaches and the intelligent use of industrial AI are required to meet the demands of modern industry. Unlike industrial AI models, general AI models are trained based on extensive system data that often does not cover the entire spectrum of possible operations. This is because general AI models do not consider conditions for different purposes (e.g., safety, design) or conditions dictated by physical and chemical laws.
Industrial AI refines the quality assurance process by automating it and detecting defects early. This increases the overall production and product quality.
Industrial AI enables the redesign of work processes and the creation of new business models based on data-driven innovation.
Applying industrial AI optimizes energy consumption, efficient use of materials, reduces waste, and lowers costs. Additionally, it enables the strategic allocation of shared resources to critical tasks.
Industrial AI enables seamless integration and analysis of data from production and process operations. This allows for informed decisions to improve both product quality and production line efficiency.
The key to successful industrial AI application lies in transforming raw data into intelligent insights for quick decision-making. From the intricacies of data management and integration to the complexities of adapting AI models to real-world production environments, manufacturers must proactively address the following challenges.
Although the industrial data environment today is a big data environment, there is a mix of structured and unstructured data that can be of poor quality (e.g., unbalanced data, missing data points, inaccurate sensor measurements, data drift, inconsistent formats, limited scope, etc.).
The credibility of AI systems in the industry can be compromised if the accuracy is not nearly perfect, as these systems could address critical safety, reliability, and operational issues. Any AI failure could have negative economic and/or safety impacts and deter the use of AI systems. By adhering to the requirements for trustworthy AI (i.e., following “The Assessment List for Trustworthy Artificial Intelligence” (ALTAI) [5]), data analysis results become comprehensible (including interpretability and transparency).
Production processes require quick decisions, and the produced workpieces can be expensive, so AI applications must respond quickly to avoid waste and other consequences. Unlike other AI systems (e.g., recommendation systems), industrial AI systems also require a very low tolerance for false positives and negatives to be used in production.
Incorporating expertise is a must to make the distinction between general AI and industrial AI clear. Data engineers and data scientists need to collaborate with domain experts and incorporate expertise into the modeling process. To maximize the inclusion of expertise, the developed models must learn adaptively and accumulate experts’ insights as knowledge.
As we have seen in the previous section, numerous challenges often result in considerable time spent before meaningful results from production are available. Sometimes this goal is not achieved because it is too complex and focus is lost. Therefore, RISC Software GmbH pursues an agile approach (with strong involvement of the actual actors) in AI-based data analysis in the industrial sector.
Figure 2 illustrates the proposed agile workflow suitable for an industrial AI project. The workflow begins with continuous discourse between industrial stakeholders and the team developing the AI solution (referred to as the AI team in the workflow). After detailed analysis by the participants, the use cases and their requirements are defined. The AI team designs appropriate solutions for the use cases after further dialogue with the industrial stakeholders. To model the appropriate data environment, the AI team creates a series of questionnaires that data providers must complete. These questionnaires form the basis for the requirements for data preparation and integration. Subsequently, the data providers supply various types of data based on the use case requirements. The AI team’s data engineers process, transform, and load this data into a data lake. This data engineering process relies not only on contributions from the AI team’s data scientists but also on contributions from industrial stakeholders, such as domain experts or process engineers. The data available in the data lake is thoroughly cleaned and prepared for analysis by the AI team’s data scientists. They perform exploratory data analysis and work with domain experts to further refine the data analysis process. The preprocessed data is then used to train AI models according to the use case specifications. The results of these AI models are then reviewed together with domain experts to make them suitable for production.
The crucial aspect of this approach lies in the active involvement of domain experts throughout the design and implementation cycle of the AI solutions.
Figure 2: Agile workflow for industrial AI solutions
Developing AI solutions valuable for manufacturing processes requires conscious enrichment with specific industry expertise [6]. This is crucial for deriving benefits from AI. Industrial AI achieves this by combining data science, AI, and industrial expertise. Within a systematic industrial AI workflow, machine learning algorithms are developed, implemented, and deployed tailored to specific industrial applications.
[1] Lázaro, O. et al.: “Model-Based Engineering and Semantic Interoperability for Trusted Digital Twins Big Data Connection Across the Product Lifecycle”. In: Curry, E., Auer, S., Berre, A.J., Metzger, A., Perez, M.S., Zillner, S. (eds) Technologies and Applications for Big Data Value. Springer, 2022.
[2] Deloitte: “AI Enablement on the Way to Smart Manufacturing”, Deloitte Survey on AI Adoption in Manufacturing, 2020.
[3] Ducic, N. et al.: “Casting Process Improvement by the Application of Artificial Intelligence”, In Appl. Sci. 2022, 12, 3264. https://doi.org/10.3390/app12073264.
[4] European Commission: Horizon Europe – Work Programme 2023-2024, Digital, Industry and Space. European Commission Decision C(2023) 2178 of 31 March 2023.
[5] High-Level Expert Group on Artificial Intelligence (AI HLEG): The Assessment List for Trustworthy Artificial Intelligence (ALTAI), July 2020, Ethics guidelines for trustworthy AI | Shaping Europe’s digital future (europa.eu).
[6] AspenTech: “The future starts with Industrial AI”, MIT Technology Review, https://www.technologyreview.com/2021/06/28/1026960/the-future-starts-with-industrial-ai/, 2021.
Data Science Project Manager & Researcher
Data Scientist
