Not having a clear goal or outcome

​One of the reasons why DX in the manufacturing industry is not progressing is that projects are carried out without clearly defining the objectives and outcomes.

DX projects are difficult due to the complexity of consensus building, resistance to change, and the difficulty of seeing the return on investment in the short term. Furthermore, because there is no precedent within the company, it is difficult to set appropriate quantitative results and evaluation methods.

As a result, projects progress without clearly defining objectives and outcomes, and people end up feeling that "there is no goal to begin with, so we don't know how to make progress," or "we're not seeing any results."

The budget available for DX is limited

Another factor slowing progress in digital transformation in the manufacturing industry is the difficulty in getting a concrete image of the costs involved. Terms like AI, IoT, and system renewal often lead people to believe that large-scale, expensive investments are required, making it difficult to secure the budget.

In addition, DX budgets are not secured continuously every fiscal year, and many companies require approval each time, meaning that budgets are often cut off if there is no sense of short-term results or effectiveness.

Furthermore, in small and medium-sized manufacturing companies, capital investment tends to take priority, and budgets for IT and digital fields can be put on the back burner. As a result, DX is often considered but not implemented.

Lack of IT personnel and promoters

Another major challenge facing DX in the manufacturing industry is the lack of IT personnel and dedicated staff to promote DX.

​DX is not progressing because there are no people in the company who are knowledgeable about digital and IT and can lead DX.

In some companies, the number of people in charge is small, and information systems managers and on-site managers handle the process while also handling their existing duties. However, this puts a heavy workload on promoters, preventing them from concentrating on promoting DX, and DX does not progress as expected.

Data infrastructure facilities are difficult to install

In the manufacturing industry, there are many legacy systems and equipment that have been in use for many years, making it difficult to establish a foundation for acquiring and integrating data. Each piece of equipment has different specifications and communication methods, and data is often fragmented.

Furthermore, craftsmanship, supported by the experience and intuition of experts, is deeply rooted in business operations, and the lack of visualization of work procedures and decisions as data is also a barrier to promoting DX. As a result, many companies stumble even before they can utilize data, and are unable to take the first step toward DX.

The growing concerns of AI/DX promoters

First, Haga  "I feel that in the manufacturing industry, thinking about AI has changed dramatically over the past few years. As a result, I think the scope of the role required of DX personnel is expanding," he says.

When you are assigned to utilize AI and promote DX, you are naturally expected to produce results.

Furthermore, with numerous projects underway, decisions must be made based on factors such as cost, effectiveness, and balance. Furthermore, once a project is fully launched, tasks that were previously not emphasized, such as interdepartmental collaboration and streamlining of work flows, begin to pile up. Meanwhile, there are very few issues that the DX promotion department can solve on its own.

In this situation, what are the barriers that DX promoters face in achieving results through DX, and how can they overcome these obstacles?

The role required of the DX promotion department continues to increase

The keywords are "vision, collaboration, external utilization, and development"

"One of the common stumbling blocks in DX projects is not having a clear vision or roadmap. As a result, there are many cases where projects end up stalling midway because they lose sight of their objectives and priorities.

"In many projects, the division of roles between internal and external stakeholders is unclear, preventing smooth collaboration. There are also cases where projects don't get off the ground smoothly due to a lack of personnel with the necessary knowledge and skills to advance DX," says Haga.

"To overcome the barriers that hinder digital transformation, it is necessary to coordinate and integrate projects and departments that are running separately, and to create an overall plan.

There are three possible solutions: develop human resources in-house, utilize external partners, or do both. It is quite rare to get by with in-house human resources alone, and in most cases, the knowledge of partners with specialized knowledge is utilized.

However, most companies first need personnel who can converse at the same level as partners, so they are also working on developing personnel at the same time." (Haga)

The risk of collecting data without analysis

"You often hear stories of sensors being installed and data being collected, but the data not being at a level that can be analyzed. There's no problem if you collect data on the assumption that it may be unusable and then take an agile approach to course correction, but the longer you wait to realize this, the greater the impact it will have on the project. With sensing, the idea that it's okay to just collect data doesn't work," says Haga.

High difficulty in collecting equipment operation data

"This means that collecting equipment operation data is more difficult than expected. You might think it's natural to be able to obtain information about the status of existing equipment, such as operational information. Management, who are particularly far from the site, tend to think that collecting data isn't difficult. This is because there is an inherent desire to avoid spending money and time on the'observation'stage, but in reality, there are many cases where issues arise." (Haga)

Lack of definition of data structure with an eye to utilization

“It is important to firmly define the storage format of the data that will be output as a data mart. , It is important to create a system with an image of the number of digits in the lot number, manufacturing model number, etc. Since the resolution of the data in the process is often different, this definition requires experience and sense. (Haga)

Continuous operation and monitoring required for AI systems

"Generally, IT systems are designed and developed according to requirements definitions, and are guaranteed in accordance with those definitions. However, when AI models are included, the story changes a little. Systems are built on the assumption that the AI will function as expected, but guaranteeing the accuracy of AI is difficult. Therefore, it is necessary to monitor the accuracy of the AI and data drift to ensure that the AI is behaving as expected." (Haga)

Factory data in particular is complex, and various factors can affect and change the trends in the data. These unexpected changes can also change the answers that AI derives, so by the time you realize something is completely wrong, it's already too late. For this reason, it's important to carefully track changes.

In addition, to respond to changes, it is necessary to have a system for managing versions of AI models and applying the most appropriate model. Without this system, no matter how good the system's judgments were when it was created, its accuracy will simply deteriorate over time.

It is also true that recent advances in AI technology have reduced the burden of such operations.
For example, with the emergence of AutoML model retraining, the development of MLOps tools, and even mechanisms for automatically detecting data drift, the environment for keeping AI systems running is steadily coming into place.
However, this does not mean that the operation and management of AI systems is no longer necessary. Rather, it should be seen as an increase in options to make tasks that have previously been performed manually easier to carry out and continue.

Current situation analysis is important for advanced production planning and labor saving

Regarding "advanced production planning and labor-saving," there is a growing need for advanced production planning to respond to sudden fluctuations in demand and small-lot production of a wide variety of products, as well as digital transformation and the use of AI to reduce and standardize the number of personnel involved in planning work.

"When creating optimal production plans, what​ ​data to look at and how to make decisions vary from person to person. To standardize this work, which is dependent on individual skills, it is necessary to visualize the current business process and create decision-making rules. Understanding the situation on-site is a more important process than developing AI models or improving accuracy." (Haga)

Current situation analysis is an important process, but the resolution tends to be rough, and it may not be used as a basis for introducing new systems and mechanisms.

The reasons for this include the heavy burden on those involved in the survey and concerns about the reliability of the survey results.Since efficiency analysis using simulators is also effective for analyzing the current situation and redesigning business processes, it is a good idea to consider this before introducing AI.

Quality improvement and standardization of personalized processes are considered together

It is best to consider both "quality improvement" and "standardization of personalized processes" at the same time, rather than thinking about them separately.

Quality control is a source of competitiveness, and many companies invest large amounts of human resources in it. However, because quality control often relies on the know-how of experienced workers, an increasing number of companies are becoming aware of the risks to its continued maintenance and development.

"With regard to quality improvement, themes such as defect causes, quality prediction, and optimal control can be raised. For example, we can analyze the causes of defects and use this knowledge in manufacturing know-how, predict the results to reduce inspection man-hours, and derive optimal parameters to perform appropriate control.

Some companies are considering automated operations that ultimately feed optimal settings back to machines, but currently the mainstream approach is to build systems that assume coexistence with humans." (Haga)

Haga concluded by saying, "I imagine that the missions and projects that you are all taking on are not all easy. I hope that by avoiding common pitfalls and referring to success patterns, you can increase your chances of success, even if just a little."