AI Predictive Maintenance for Equipment: The Key to Avoiding Failure

Table of Contents

1. Introduction
2. Challenges in Manufacturing Sites and Emerging Limits of Preventive Maintenance
3. AI-Based Preventive Maintenance (Predictive Maintenance) in Focus and Points to Note for Implementation
4. The Role and Bottlenecks of Annotation in Preventive Maintenance
5. The Role of Proxy Services Supporting Annotations Requiring Skilled Techniques
6. Summary
7. Features of Human Science Annotation Services

1. Introduction

Preventive maintenance in manufacturing has been a challenge addressed for many years, even before the advent of AI and IoT.
Responding to maintenance and repairs only after equipment stops due to failure not only greatly impacts production plans, quality, and delivery schedules, but equipment downtime also causes various losses that are strongly disliked on the manufacturing floor. Therefore, on manufacturing sites, participatory preventive maintenance activities, such as those represented by TPM (Total Productive Maintenance), have long been practiced and have yielded results. However, in recent years, the limitations of these approaches have begun to surface, and efforts to implement AI have started to advance. This time, we would like to explain the recently emerging limits of human-based preventive maintenance, the benefits and precautions of AI implementation, and how to work with AI.

2. Challenges in Manufacturing Sites and Emerging Limits of Preventive Maintenance

In recent manufacturing sites, issues have become apparent such as the decline of skilled workers who supported preventive maintenance and the shortage of young personnel, resulting in tacit knowledge not being formalized and passed on. Furthermore, even when work is handed over from skilled workers, inspections tend to become mere "check tasks," where inspections can be performed but judgments cannot be made. In addition, with the advancement and increasing complexity of equipment, signs of abnormalities whose causal relationships cannot be seen through human experience cannot be standardized, and as a result, sudden equipment stoppages have been increasingly observed.

Even if daily and periodic inspection items for equipment are formalized and passed down through manuals and checklists, in highly challenging inspection areas, the judgment results for the same inspection items can subtly differ depending on the person in charge. This is because it was precisely the skilled workers’ senses, cultivated through many years of experience, and their understanding of the quirks unique to each piece of equipment that made accurate judgment possible. In other words, it was possible because they possessed tacit knowledge that is extremely difficult to formalize. With the decline of skilled workers and the younger generation inheriting their roles, the limits of preventive maintenance are also beginning to emerge.

3. AI-Based Preventive Maintenance (Predictive Maintenance) in Focus and Points to Note for Implementation

AI is expected to be an effective means of passing on the previously unseen judgments of skilled technicians to the next generation and supporting people, and its development and implementation are currently progressing. Predictive and preventive maintenance is no exception, and efforts to implement AI are advancing in various companies as a solution to the challenges faced on-site. Continuous monitoring by sensors, anomaly detection algorithms, and other technologies now make it possible to continuously watch over equipment conditions without constant human presence. As a result, the risk of sudden equipment stoppages is reduced, early detection of small changes and signs often missed by human inspections is enabled, and the decline in skilled workers is supplemented. AI-driven preventive maintenance also helps reduce excessive inspections and parts replacements, prioritize maintenance tasks, and optimize maintenance costs and workload. Many successful cases of AI implementation have been observed as a complementary means to maintain stable operations with limited personnel.

However, it should be noted that simply introducing preventive/predictive maintenance AI does not complete the process. In actual workplaces, the way equipment is used, operating conditions, installation environment, and maintenance policies (such as where to define an abnormality) all vary. Therefore, even if the introduction of such a generic AI initially proceeds smoothly, when changes occur in the above conditions, the reliability of alerts issued by the AI can suddenly decline, making it impractical in many cases. Just as with any AI, it becomes necessary to incorporate your company's unique skilled workers' know-how into the AI's decision criteria and update it according to the situation, and often it is also necessary for your company to carry out these updates internally.

In other words, to ensure that preventive/predictive maintenance AI remains practically usable on an ongoing basis, it is necessary to update the judgments of skilled workers as shown below according to the equipment usage, operating conditions, installation environment, and maintenance policies, prepare AI training data tailored to each situation, and perform additional learning.

● Which states are called "abnormal"
● Which changes are regarded as "signs"
● At which stage should action be taken

What is necessary for this is the annotation work that incorporates the experience and tacit knowledge of skilled workers into data.
Annotation is not just a simple labeling task; it involves making judgments such as "why this state is abnormal" and "from what point changes in data fluctuations should be considered warning signs," and categorizing them, for example, as "normal," "caution," or "abnormal signs," then labeling the data accordingly.

4. The Role and Bottlenecks of Annotation in Preventive Maintenance

There are cases where the preventive/predictive maintenance AI provided by equipment suppliers does not apply, but ultimately, to develop or purchase and nurture AI that is practical and suited to your own equipment and usage, high-quality and relatively large amounts of training data embedded with the skills and experience of skilled workers are indispensable. However, in reality, even with the same equipment, usage and settings differ, standards change depending on the line or process, and there are challenges such as limited accumulated past data and failure data, making data preparation not easy. Additionally, it is necessary to embed judgments filled with the experience and know-how of skilled workers into the data, but since the skills and experience of skilled workers are often not verbalized and remain ambiguous, if the judgment criteria held by skilled workers cannot be fully translated into annotation work specifications and the work proceeds, the AI may ultimately fail to reproduce the "sense of skilled workers."

Not only the work specifications but also the annotation work stages require the incorporation of skilled techniques; however, having your own skilled workers perform a large volume of annotation work involves various bottlenecks such as securing costs and work time.

5. The Role of Proxy Services Supporting Annotations Requiring Skilled Techniques

If it is difficult to have your own skilled workers perform annotation or prepare training data, outsourcing to an annotation vendor can be considered. However, while annotation vendors can be expected to secure a large amount of training data in a relatively short period, they rarely possess specialized technical knowledge of the relevant equipment, etc.

However, many experienced annotation vendors, through communication with customers and repeated feedback on the work, absorb the ambiguous nuances that are difficult to verbalize and the tacit knowledge possessed by skilled workers, which inevitably exist in any annotation. They verbalize and standardize what can be verbalized, and by deploying and thoroughly implementing these in the optimal way to the workers, they possess the know-how, experience, and key points to stabilize accuracy and quality. By utilizing the outsourcing and agency services of annotation vendors who leverage such processes as their strength, it is possible to reproduce skilled workers' techniques with considerable accuracy. Additionally, parts of the skilled workers' techniques that are difficult to take outside the company or convey can be handled in-house, while the rest can be entrusted to annotation vendors, allowing for a division of labor that balances both the quality and quantity of data.

Even knowledge and experience of skilled workers that may seem relatively easy are difficult and unrealistic to fully verbalize in advance and document in specifications. To pass on such knowledge and experience, it is necessary to divide the work into smaller parts for partial or sample deliveries to obtain feedback from the customer's skilled workers, as mentioned above, and to hold meetings or similar occasions to verbally align on judgment criteria and subtle nuances that are difficult to verbalize. Therefore, simply faithfully executing annotation work according to verbalized specifications is insufficient as an outsourcing service.

Depending on the situation, it may be necessary to work as dispatched staff under the customer's direction, considering the required security level and the intensity of communication. In other words, it can be said that only annotation vendors who can closely collaborate and communicate with customers, provide flexible solutions and responses while being supportive, and offer such services can achieve and handle these requirements.

6. Summary

As mentioned so far, preventive/predictive maintenance AI is continuously being developed for the transmission of skilled workers' techniques, and recently it has been expected as one of the solutions to challenges such as labor shortages faced by the manufacturing industry, including smart factory initiatives. On the other hand, simply introducing such general-purpose AI does not solve all problems or complete the process; like other AIs, if usage conditions change, additional training and maintenance of the AI itself will be necessary. Furthermore, the training data required for AI learning must be high-precision, relatively large in volume, and diverse, incorporating the know-how and experience of skilled workers.

By considering these realities when introducing AI, the implementation of AI that inherits the experience and know-how of skilled workers to the next generation and within the company will be successful.

7. Human Science Teacher Data Creation, LLM RAG Data Structuring Outsourcing Service

Over 48 million pieces of training data created

At Human Science, we are involved in AI model development projects across various industries, starting with natural language processing and extending to medical support, automotive, IT, manufacturing, and construction, just to name a few. Through direct business with many companies, including GAFAM, we have provided over 48 million pieces of high-quality training data. No matter the industry, our team of 150 annotators is prepared to accommodate various types of annotation, data labeling, and data structuring, from small-scale projects to big long-term projects.

Resource management without crowdsourcing

At Human Science, we do not use crowdsourcing. Instead, projects are handled by personnel who are contracted with us directly. Based on a solid understanding of each member's practical experience and their evaluations from previous projects, we form teams that can deliver maximum performance.

Generative AI LLM Dataset Creation and Structuring, Also Supporting "Manual Creation and Maintenance Optimized for AI"

In addition to labeling for data organization and creating training data for identification-based AI, we also support structuring document data for generative AI and LLM RAG construction. Since our founding, manual production has been our main business and service, and we now also assist with organizing business knowledge and manual creation aimed at future generative AI and RAG implementation and utilization. We provide optimal solutions leveraging our unique expertise in the structure of various documents.

Secure room available on-site

Within our Shinjuku office at Human Science, we have secure rooms that meet ISMS standards. Therefore, we can guarantee security, even for projects that include highly confidential data. We consider the preservation of confidentiality to be extremely important for all projects. When working remotely as well, our information security management system has received high praise from clients, because not only do we implement hardware measures, we continuously provide security training to our personnel.

In-house Support

We provide staffing services for annotation-experienced personnel and project managers tailored to your tasks and situation. It is also possible to organize a team stationed at your site. Additionally, we support the training of your operators and project managers, assist in selecting tools suited to your circumstances, and help build optimal processes such as automation and work methods to improve quality and productivity. We are here to support your challenges related to annotation and data labeling.