GenAI: a hype or a real benefit for manufacturing industries?
Generative AI has been a hot topic since the end of 2022 with the advent of ChatGPT promising to ease our daily work by providing out-of-the-box analysis and studies. Nevertheless, recent history has also shown that GenAI provides strong opportunities as well as some limitations and risks that must be addressed.
Before we have a look into the opportunities that GenAI might provide for manufacturing industries, we need to define what GenAI is.
GenAI is a type of AI that creates new data from the existing data that it has been trained on.
It is based on large models (foundation models) that are pretrained on comprehensive amounts of data. Leveraging algorithms can enable revolutionary conversational user assistance, and spur advances in content creation, software development and product ideation.
Can GenAI generate comprehensive benefits for manufacturing?
With fewer than two years of broad availability, the use of GenAI in manufacturing is promising, but still in the early stages. According to research conducted by IDC in early 2023[1], 27% of the surveyed companies reported that they are already investing in this technology. Additionally, 41% of the respondents believe that GenAI can impact manufacturing. That is significantly higher than the impact the respondents attributed to product development (28%) and sales (20%).
GenAI offers opportunities across the manufacturing value chain, from R&D to production, sales and services. The most promising use cases for AI are currently in sales and services (e.g., sales automation, personalized offerings) and in R&D (e.g., software development, product ideation).
The focus of this post is on emerging use cases in manufacturing execution, where AI is poised to significantly improve Overall Equipment Efficiency (OEE) and address the impending brain drain in production. This skills gap is a result of the baby boomer generation reaching retirement age and a decrease in the number of skilled personnel entering the workforce.
GenAI use case 1: plugging the production brain drain
To improve the skills gap, digitalization and the capture of historical data are crucial. This includes machine data, maintenance and repair records, equipment manuals and even external manufacturer data.
By feeding this data into a foundation model, Manufacturers can gain valuable insights into potential equipment failures, necessary adjustments, required maintenance schedules and spare parts needs. This proactive approach can significantly prevent downtime.
GenAI models offer a wider range of benefits including predictive maintenance, improved quality control, optimized production planning and efficient inventory control.
GenAI use case 2: inventory control
By simulating various production scenarios and predicting demand fluctuations, GenAI models can significantly reduce inventory levels while ensuring resource availability and considering demand fluctuations.
GenAI use case 3: quality control
GenAI can revolutionize quality control by learning from vast datasets of past product images. By identifying defects and using this knowledge to build a model, manufacturers can predict the failure rate of future products.
Machine learning algorithms play a critical role in predicting equipment failures within a defined timeframe. Predictive maintenance enables replacing parts before a breakdown occurs, maximizing equipment lifespan.
These three GenAI-powered use cases deliver numerous benefits for manufacturers including cost savings, reduced downtime through predictive maintenance and significant efficiency gains through process automation.
GenAI limitations to be addressed
GenAI applications offer multiple promises. However, there are major challenges to be overcome:
- GenAI outputs must be fully reliable. Therefore, measures need to be taken to assure reproducibility and repeatability.
- GenAI bias can be a significant problem. A major source of bias can be the data used to train the underlying data models, as well as the algorithms and techniques applied. Bias can also be inherent in human generated data applied to the training of the models.
- Bias in the dataset can also lead to generated content that amplifies this bias. One consequence can be the generation of misleading content. Acceptance and credibility of the technology depends on the mitigation of such biases.
Three approaches to mitigation
- It is important to establish feedback-loop procedures to include (human) subject matter experts who review GenAI results. Reviews should be based on the implementation of appropriate evaluation metrics to evaluate the quality and reliability of the AI-generated content.
- Data quality is paramount in defining datasets and models. Therefore, adequate corporate data management needs to be established.
- Finally, GenAI requires skilled resources that are sometimes difficult to find in a labor market characterized by a shortage of skilled personnel. The respondents to the IDC study mentioned earlier pointed to the lack of skill as the major drawback for the adoption of GenAI (35%) followed by insufficient tools, and issues with data availability and quality (28% each). These challenges are often countered by pooling scarce resources in a corporate center of excellence for AI, including a strategy to acquire and educate talent.
Furthermore, the support of external consultants and AI providers will be needed to fully explore the benefits of GenAI.
GenAI is still young and full of potential for the manufacturing industry. However, it is not without limitations and serious consequences. In time, the potential benefits will outpace the current limitations encountered.
To support manufacturing organizations in fully exploiting, scaling, and leveraging the transformative power of Generative AI, Eviden has launched an ambitious Generative AI Acceleration program.
References and sources
- [1] IDC: The state of manufacturing and generative AI adoption in manufacturing organizations, 1Q23 (respondents from 16 countries and companies > 500 employees)
