YANMAR Technical Review

Abstract

The Yanmar Group is transforming itself into a company that creates customer value, providing products that address customer challenges in this era of volatility, uncertainty, complexity, and ambiguity (VUCA) where traditional values and business models are no longer effective. Yanmar Energy System (YES) is dedicated to delivering daily peace of mind to customers by utilizing its Remote Energy Support System (RESS) for the remote collection of big data from YES products. This article describes how digital solutions using remote data are being combined with rapidly advancing digital technologies to effectively address customer challenges.

1. Introduction

Rather than seeing itself solely as a manufacturer, the Yanmar Group is transforming itself into a company that creates customer value through the supply of products and solutions that resolve customer challenges. Yanmar Energy System (YES) is the Yanmar Group company responsible for its energy business. As a pioneer in distributed electricity generation and air conditioning, it is contributing to the creation of a sustainable society and circular economy through the supply of reliable and highly economical energy systems. The Remote Energy Support System (RESS) started out in 1984 as a system for monitoring an emergency generator on a remote island and is now used to watch over customer sites throughout Japan, including gas heat pumps and cogeneration systems. A contact center was opened in 2020 to deliver services that prevent interruptions to customer operation, featuring the collation on RESS of big data obtained by the 24-hour/365-day remote collection of information on the operation of YES products from sites around the country.
YES advocates practices that create customer value by speeding up the process of working through the cycle of listening to what customers have to say, identifying latent issues, proposing solutions, evaluating benefits, and gathering feedback (Fig. 1). This involves utilizing the remote data collected on RESS to implement ways of resolving ever more diverse customer challenges through the use of advanced digital technologies (including sensing, the IoT, and AI). This article describes how practices that create customer value are utilized in the development of digital solutions that combine remote data and digital technology to overcome challenges.

2. Development Background

The “job chain” is defined as the sequence of steps from a customer purchasing a product through purchasing its replacement (Fig. 2). The following three key issues from this job chain were identified by listening to what customers have to say.
2.1. Routine inspection of equipment relies on human labor and judgment
2.2. While customers want their equipment to last, maintenance costs are expensive
2.3. Customers are concerned about unexpected downtime

2.1. Routine inspection of equipment relies on human labor and judgment

Given Japan’s shrinking workforce and the aging of skilled craftsman, customers want inspections to be highly accurate, and to be done without relying on human experience and with the lowest possible man-hours and cost. Unfortunately, current inspection practices require that the inspector spend a lot of time both on inspecting and on maintaining inspection records, having to use their own senses to read meters, perform equipment checks at designated points, and enter the results manually on paper forms. Inspection quality also tends to vary, with the quality of work being dependent on the skill and experience of the person doing the inspection.

2.2. While customers want their equipment to last, maintenance costs are expensive

Customers want to maintain reliable operation at the optimal maintenance cost, keeping their equipment in operation for a long time and avoiding downtime. At present, this usually involves time-based maintenance (TBM) undertaken routinely with part replacement times set by the manufacturer based on assumptions about design life, past failure rates, and how the equipment is used by the customer. Unfortunately, the timing of part maintenance in TBM is determined beforehand rather than based on part condition, and while this ensures that the equipment continues to operate reliably, there is the potential for over-maintenance if the customer’s use of the equipment puts less stress on it than the manufacturer has allowed for.

2.3. Customers are concerned about unexpected downtime

If, on the other hand, a customer puts equipment to harsher use than the manufacturer anticipates, there is the potential for unexpected downtime due to a part failing before it becomes due for replacement. Here, “downtime” refers to time when the equipment is not available for use. Not only does such downtime require unplanned part replacement or repair work, it also risks bringing customer operations to a halt due to the equipment being temporarily out of service.

3. Digital Solutions from Yanmar

Yanmar has developed the following digital solutions that address each of the three key customer issues described above.
3.1. Total equipment visibility
3.2. Condition-based maintenance (CBM) using machine learning
3.3. Predictive maintenance using fault prediction models

3.1. Total equipment visibility

Total equipment visibility involves making information about equipment condition available by monitoring it using sensors and quantifying the information this generates (such as temperature, pressure, and humidity) (Fig. 3). The need for an inspector to check meters or write down data on an inspection form by hand is eliminated by making data available from a wide variety of sensors. This includes the real-time engine data shown on the top row of Fig. 3 and the real-time audio, vibration, and camera feeds shown on the bottom row. This means that inspections can be completed with minimal time and effort. Furthermore, by quantifying subjective inspections performed using human senses (such as for leaks or abnormal sound or vibration), doing things this way frees inspections from being dependent on human experience.
One way to achieve this is to replace the human senses of sight, hearing, and touch with cameras, microphones, vibration sensors, and temperature sensors, with IoT devices used to collect this data at regular intervals and upload it to the cloud (Fig. 4). Installation is simplified and the need for extensive reworking of existing plant is avoided by using wireless communications to link the sensors and IoT devices. Likewise, the IoT devices upload data to the cloud using secure communications. Yanmar has also built a data infrastructure platform on the cloud with data storage, analysis, and presentation functions. While the existing RESS only monitors Yanmar engine generators, the new system also supports engine generators from other vendors and equipment such as radiators and pumps, thereby providing solutions to an even wider range of customer issues with more efficient maintenance for all equipment.
When the digital solution was tested at customer premises, the feedback from people stationed at remote locations away from the equipment site was that the solution made it easy to check on the status of equipment at any time. The feedback from people stationed on-site, meanwhile, was that it enabled anyone to perform maintenance at the same level without devoting a lot of time to the work. Figures from the testing indicate that the solution reduces the amount of time spent on routine inspection work by about 70% compared to past practice.

3.2. Condition-based maintenance (CBM) using machine learning

Whereas TBM involves performing routine maintenance based on the part replacement times recommended by the manufacturer, CBM monitors equipment in real time and schedules maintenance based on its actual condition. As this enables equipment to be operated at the optimal maintenance cost, CBM represents a new generation in maintenance practices.
In this instance, adopting CBM for a diesel generator involved two main steps (Fig. 5). The first was to analyze equipment data (parameters such as exhaust temperature and cooling water pressure) using an AI algorithm. As very little data is available on faults, this step used unsupervised learning, a machine learning technique that can train a model using data for normal operation only. To detect signs of potential problems, the extent to which real-time data diverges from normal is calculated by comparing it against the model trained using only data for normal operation. If the degree of divergence is large, this is judged to be anomalous. By doing so, the system can estimate the state of the generator, such as whether the exhaust temperature is higher than usual or the intake air temperature is lower, for example. As large amounts of normal operation data are used for training and inference on an AI that was not included in the existing RESS, it is possible to extract patterns and trends from the data in minute detail. This has made it possible to detect signs of potential problems that a human would fail to notice. The second step uses the detected warning signs and a fault diagnosis map to deduce whether the diesel engine or other systems have a problem. This fault diagnosis map consolidates more than a century of expertise in diesel engines that Yanmar has accumulated since its founding, linking each engine part to the abnormal sensor readings they can cause. Use of this proprietary knowledge has enabled part condition to be estimated with high accuracy.
The feedback when a customer trialed the system was that, while the annual maintenance cost is high, it allows equipment to be operated with optimal maintenance costs. It was estimated that the system could reduce the cost of spare parts by about 20%.

3.3. Predictive maintenance using fault prediction models

The CBM practices described above allow equipment to be operated with optimal maintenance costs by continuing to use consumable parts for as long as they are in good condition. Predictive maintenance, on the other hand, is targeted at parts other than these routinely replaced parts and works by analyzing equipment data to predict potential failures and schedule parts replacement accordingly. As this reduces downtime by having appropriate repairs performed at an earlier timing, it helps companies keep their operations running safely and reliably.
In this instance, predictive maintenance was applied to gas heat pumps (GHPs), with prediction models being built for a number of parts for which fault prediction has proved difficult in the past. This was done by combining machine learning and data on GHP internal operation (equipment design expertise and data used for internal control) with remotely collected data on past GHP operation. The prediction models could be used to predict the normal values for data that would be influenced by a fault. A fault in the target part could then be predicted by checking the actual values against these predicted normal values (Fig. 6).
Use of the digital solution improved the percentage of accurate fault predictions for applicable GHP parts from 31% to 74%.

4. Conclusions

The work described in this article has helped to successfully overcome challenges by focusing on the predicted customer job chain and using remotely collected data and digital technology for total equipment visibility, CBM using machine learning, and predictive maintenance using fault prediction models.
In the future, Yanmar intends to continue with the development of solutions that its customers cannot live without, expanding the scope of these newly developed digital solutions and resolving the challenges that customers face in their job chains amid a rapidly changing world.