Posted inSectors

Guest column: PoC is not enough

AI in manufacturing often fail to go beyond PoC and the reason for this goes beyond lack of budget or expertise.

Sandeep Pendharkar, Sr. Director and GM, HPE Services Gu

The Indian Artificial Intelligence market is valued at USD 7.8 billion as of July – August 2021. This represents a 22% increase in size of market over 2020. AI plays a significant role in the manufacturing industry. It helps in better product development, quality improvement and also leads to market adoption.

The introduction of AI in production often fails during transition from proof of concept to regular operation. This can only be avoided with a holistic approach that takes into account the entire spectrum of commercial, technical and organizational dependencies. 

Companies underestimate the systematic AI challenges

However, several years and thousands of press articles and analyst reports later, the situation can seem somewhat disillusioning. While overall AI adoption continues to increase – according to McKinseys “State of AI in 2021” – manufacturing is still far behind. A key reason for that is that a large number of AI projects do not go beyond a test phase, the so-called proof of concept (PoC). The causes for this failure lie deeper than, for example, a lack of expertise or budget – in many cases manufacturing companies underestimate the systematic challenges of introducing AI.

The way in which the PoCs are set up is only a symptom of this. They usually take place in a protected environment and are focused on the application and training of the AI ​​models with data – but the need to integrate the AI ​​solution into the existing information and production technology and its processes is often neglected. This includes, for example, the life cycle management of applications and data, security, operational planning and control processes and operational safety. As a result, the PoC does not provide serious evidence of technical feasibility, nor can it be used to calculate a solid business case.

A holistic approach to introducing AI

As boring and tedious as it may sound, the introduction of AI in manufacturing can only be successful with a holistic approach. The PoC should be the tip of the iceberg – the outcome of a series of underlying decisions and projects, where initiatives are derived from strategies that are implemented through technical, organizational and cultural transformation activities. 

A holistic approach to the introduction of AI in manufacturing includes the following aspects, among others.

Value creation – benefit and cost analysis

The added value when using AI is created through information, insights and (autonomous) actions and processes derived from them. The available data is the foundation for this – however data does not necessarily become useful information only by applying AI. They only become that if they are processed in a specific context and for a specific purpose. Value creation analysis on the one hand evaluates the benefits of the information obtained with the help of AI; on the other hand, it determines the quality of the data and the effort for data acquisition and processing as well as the associated investments for operational production – including process, technology and personnel costs. The result is the business model or business case.

Process – development and lifecycle of the AI ​​application

If the value analysis comes to a positive result, the development and introduction of the AI ​​application begins. This should follow a DevOps philosophy in which all relevant teams from production and operations as well as AI and IT experts work together (in the AI ​​context, this is also referred to as MLOps and DataOps). This ensures that the integration into the IT and manufacturing processes is taken into account right from the start. 

In India, large scale applications of AI are being trialled everyday across sectors. In Uttar Pradesh, for example, 1,100 CCTV cameras would raise an alert when the crowd density exceeded a threshold, and the connected Integrated Command and Control Centres provided the security authorities with relevant information. NIRAMAI, a start-up, has developed an early-stage breast cancer detection system using a portable, non-invasive, non-contact AI-based device. 

Architecture – decouple data from the applications

The processes described above take place in IT and manufacturing environments that are highly fragmented in many companies – that is, there is no continuous access to tools and data, processes do not match, standards and integrated security concepts are missing. Such an environment is deadly for any AI implementation.

The basis for solving this problem is the introduction of a data-centric architecture. At its core, it decouples the data from the applications that generate them by channelling them via a central data hub. Each application acts as a “producer” of data for the data hub, each query is a “consumer” of the extensive, distributed database. All of this is embedded in an overarching data governance framework.

Competencies – interdisciplinary cooperation

In many cases, AI PoCs is set up too one-dimensionally because they are carried out by data scientists who understand a lot about data and models, but less about system architecture and IT processes – and next to nothing about the processes in a factory. A successful AI introduction requires the right mix of skills from different departments in order to plan, develop, roll out and put into operation the application itself and its integration into IT and production processes. A typical team consists of roles such as business analysts, data scientists, machine learning specialists, data engineers, software engineers and finally a project manager, which can be filled by people from your own company or from a service provider.

The good news is that there is a continuous transition from automation to autonomy. The current requirement in the Manufacturing industry is the better integration of factory stack and creating a synergetic ecosystem to enable MaaS (Manufacturing as a service). With this, software defined manufacturing is evolving rapidly in India to bring next level of automation. 

AI is becoming crucial for the manufacturing business in India to stay competitive in the market – and it will soon be a much-needed accessory for many businesses. It goes without saying that the manufacturing industry will need to grow at the same pace as technological improvements to make full use of the innovations.

(Sandeep Pendharkar is the Sr. Director and GM at HPE Services)