Where a machine-tool CTO spends his engineering judgement

A company of our size cannot develop everything that a much larger competitor develops, so most portfolio decisions are about what we will not own. I divide our ¥9.7 billion of R&D into three horizons: improvements to current machine families; shared platforms such as spindles, thermal models and operator software; and a few longer research bets with university partners. Every project passes four gates, and at each one its leader presents the evidence that would make us stop as well as the evidence for going on. I track two measures. One is the share of spending on modules used by more than one machine family, which has risen from 31% to 52% since 2022. The other is the number of projects stopped at a gate, which I read as a sign of health rather than failure. Customer requests matter, but I do not let a single large order reshape a platform; that choice belongs in the shared-module plan the board sees each year.

Most of the accuracy a customer buys from us is kept or lost through heat. Spindle bearings, ball screws, motors, coolant and the workshop itself move a machine's structure by microns over a shift. My doctoral work was on compensating for that drift in grinding machines, and I still review the thermal design of every new platform. I ask for test data from a complete working day, including a cold Monday start and a long roughing cycle on a warm afternoon, rather than a laboratory figure taken after warm-up. We combine structural choices, such as symmetrical columns and controlled oil temperatures, with software compensation built from sensors on the machine. The hardest questions are partly commercial: how to state accuracy honestly in a specification, how to verify it on the customer's floor, and how much of the correction to entrust to a model rather than to cast iron.

About 23,400 of our machines in customers' plants now send health data to our service centres: alarms, spindle load and vibration, temperatures, lubrication and axis condition. I am responsible for the engineering and for the terms under which that happens. The values a machine sends are listed on its own screen, owners can switch categories off, and part programmes and workpiece measurements stay in the plant unless the customer decides otherwise. Customers who cannot permit any outside connection, including many aerospace and defence-adjacent plants, use an on-premises gateway that keeps everything inside their own network. We judge the service by first-visit fix rates and downtime avoided, not by the volume of data gathered. I work with our service division on retention and deletion, with our legal team on each market's requirements, and I report the programme's security reviews to our audit and supervisory committee once a year.

Several of our five-axis machining centres and high-precision grinding machines fall within the categories of machine tools that Japan, like other members of the Wassenaar Arrangement, controls for export, so shipments to many destinations need a licence under the Foreign Exchange and Foreign Trade Act. I do not decide licensing; our export trade control office does, and it answers to our chief compliance officer. My responsibility is to design products that make its work possible. That means classification settled at the concept review with the export office present, relocation detection on the machines that need it, configuration choices that do not depend on limits a technician could lift by editing a parameter, and service procedures that confirm where a machine is before we restore it after a move. I also make sure our engineers understand that technical assistance and software updates can raise export questions of their own.

Our accuracy still depends on about forty technicians who hand-scrape slideways and mating surfaces, and on a younger group of software engineers who did not train as machine builders. I treat both as engineering capability rather than as a production or IT matter. For scraping, we have instrumented the craft: contact patterns are imaged and scored after each pass so that an apprentice can see what a master sees, while robots take the heavy rough work on large beds. For software, every new engineer spends six weeks in assembly, the test cell and field service before writing code for a machine, and our chief software engineer signs each product release alongside the chief mechanical engineer. With our human resources division I have also introduced a technical career ladder on which a master technician or principal engineer can reach executive-level pay without managing anyone, which is still unusual in a Japanese manufacturer.