Automating Japan

Automation is a funny thing. People think automation magically happens. It does not. Why? Because automation involves the study of human processes and machines that dynamically interact with people to achieve a result. Human processes are unique to the job and the requirements of replicating them involves a complex dance between practical measures and ambitious goals.

When we think about automation, we think about replacing someone’s job, and by doing so, a company can lower costs and improve productivity. Well, that is the promise. Human processes are sometimes complex (think neurosurgeon) while others are simple (warehouse worker). According to McKinsey, 70% of jobs can be automated—but, as they say “the devil is in the details”. So, what does that mean? It means the details are in different human processes. ?

Most of the time human processes are a blend of complex and simple tasks, depending on the industry. In many cases human processes cannot be completely automated which requires machines, co bots, to work alongside people as a method to reduce physical stress, increase efficiency and improve worker safety. A co bot is a collaborative robot that works alongside people, it augments (not replaces) their work.

In the Spring of 2018 I was asked to help Mitsubishi Chemical automate their factories. At the time I ran the venture capital division of Mitsubishi Chemical and when they (C-suite) learned of my robot background they asked for my help. They wanted to know about my work at Panasonic, Omron and at Anybots, and my experience building Ugobe, my former robot company. After a number of meetings with the executive team they invited me to head up Mitsubishi Chemical’s first factory automation project. On this project I worked with two colleagues, Satoshi Yoshida and Tomiyasu Atsushi who helped with language translation, introductions to factory management teams and in understanding cultural protocols—very important in Japan. ?

The project was ambitious—to automate 1 factory line within 18 months. The project spanned 15 factories including Mitsubishi Chemical’s carbon fiber factory in Hiroshima, they’re prized facility. I toured each factory, sat down for meetings with general managers and their engineering teams. We then visited the production lines to review their processes, equipment and production teams. One factory, located in Fukuoka, Japan was a consortium of factories making special polymers and silicate, the finest in the world.

The silicate factory was fascinating, their processes involved the handling of large caldrons of silicate that needed to be precisely placed on a stone tray, which is then carefully slid into a 4000 degree oven for curing. The caldrons weighted tons, so they used a forklift to move them using 3 workers to help place the caldron perfectly on the stone tray. They wanted to automate this process. The behavioral processes played a key role here. Each worker paid attention to a different parameter—otherwise the caldron would be placed incorrectly and could crack. I was able to successfully characterize their human processes which informed my recommendations. In the end we automated the process using an autonomous forklift and 3 stereo cameras to map and guide placement of the caldron. This solution saved them $4M a year in labor costs, while improving production output and reducing losses of cracked caldrons and the respective silicate.

I visited a number of factories and after my 2nd factory visit it became apparent the chemical factories are not like anything you may be familiar with. When people think about factories (i.e. manufacturing) they typically think about electronic manufacturers like Foxconn or Flextronics. I spent time on the Foxconn campus in China (they built my robots) so I understand the layout of their factories. Well, chemical factories are large science projects. They don’t look anything like a Foxconn or Jabil factory. They are a series of pipes, mixers, reactors (furnaces) and testing stations. For chemicals to properly cure they must be kept in motion for some time, hence the pipes that carry the chemicals can be miles long in some cases. Eventually chemicals are made into pellets, slurries or into powder form. As a result the processes chemical factories implement are radically different than traditional manufacturing so I spent a lot of time observing them to understand where automation would or would not work.

Now keep in mind Mitsubishi Chemical is a chemical company, not a robot company so we didn’t have engineers on staff to build things. And given the tight schedule of 18 months I didn’t have time to hire or train workers. So, I approached SRI (Stanford Research Institute) to help design and build a robot for our application. I worked with them before at Panasonic where they helped build a few robot prototypes for me so I knew they were capable.

Once I hired SRI I received the green light to initiate the project. We kicked things off by evaluating the 15 factory use cases and the characteristics of human processes. The carbon fiber factory use case was intriguing but posed too many variables to be considered a “doable” project within 18 months. In this case, the carbon fiber strands were string thin and would fray into hairball like clumps that needed to be removed from the production line without touching the non-damaged fibers or it would create a fault code and shut everything down. Hence, too complicated but a fun project under different circumstances. ?We continued the review process until we identified the Shiga factory where Mitsubishi Chemical makes polyester film that goes into iPhone and Android screens. The Shiga, Japan factory was immense as it needed to support large volumes of smart phone customers. ?

Kawai Tomoyuki became our point person on the Shiga project. He was the technical lead for the engineering team. Kawai and Yuta Shiga provided exceptional details of their human processes and environmental constraints. This led to a series of questions which resulted in one of our 3 visits to the Shiga factory. I traveled to Shiga, Japan with the SRI team and my colleagues, Satoshi and Tomiyaso to see first-hand the production process. The meeting was eye opening. After this we visited the factory floor and reviewed the process they wanted to automate. Remember the 3 workers that were helping the silicate caldrons to be placed on stone trays (?), well the Shiga factory has 3 similar workers, but their job is completely different.

The process involves grabbing a 2 meter wide sheet of polyester film moving at 20 meters a minute, cutting it, then walking it 4 feet to the other side of the production line, inserting it into small teeth that then grab it and pulls it into a 100 foot furnace that stretches the width by 3 times. This is obviously a tricky human process. We asked them to repeat the process, recording it with our phones to re-examine it later. Out of the 5 attempts, they failed on 3 of them. The challenges became more apparent to us but we identified 4 core components that could be automated—(1) the grabbing of the film, (2) cutting of film, (3) movement of film to other side of production line, and (4) placement of film into grippers. Here is an example of what their original process looks like:

Now here is an example of what we built to automate the process:

The first image is our working prototype, built in Menlo Park by SRI. The second image is a rendered version of the proposed design. The robot grabs the film using electro-adhesion (no damage to polyester film), cuts it and hands it to an operator. We designed a fully mobile version with a robot arm to insert the polyester film into the other side of the production line, but upper management didn’t want to invest the additional $1M to build it so we settled on a stationary robot arm and our specialized endofactor. Once the robot was installed, we noticed immediate results--- the robot performed flawlessly with no errors. This was a test run, the final versions (there are 4 now) are working 24x7 with NO errors. The film was cut and handed off to one worker (reduced from 3) who then placed it in the other side of the production line.

There is always a human component to automation, and it was clear to my team and SRI that the three workers were considered important by their colleagues because of their ability to handle the polyester film. Now that we are reducing the process to 1 worker, they would have to be reassigned and perhaps lose their special status with their work mates. However, we saw an opportunity to put together a training workshop for the workers on how to use and service the robots. With that in mind we created the curriculum and schedule for the workshops so the workers could take ownership of their relationship with robots. I believe all companies that are automating their workforce should do the same.

We solved a number of technical hurdles in our project. A few of these included the electro-adhesion gripper that required additional force form a strong electrical bond with the film-- which required a few technical updates to the gripper. We also solved some camera issues around the reflective nature of the polyester film. And the Universal Robot arm would not move at the speed we needed so we eventually went to Yaskawa that didn't have a safety limitation (governor) on speed.

As a result of this project Mitsubishi Chemical claimed their first victory in their plans to automate manufacturing. And we achieved a 6-month ROI on the robots, when the company expected 18 months. The CEO thanked us for our efforts and Takayuki Aoyama, who heads up production engineering for Mitsubishi Chemical is now a believer in robots and automation... and he's become a good friend.

I learned a lot from this project. Chemical factories are large industrial complexes that span for miles. Many processes are isolated from one another through distance and regimen. The process of polymerization is not the same as co-polymerization and therefore is in a different facility—hence workers are not aware of upstream or downstream processes. Many of the worker tasks are physical (turning valves, handling large barrels of material…) and dangerous. As a result, human processes for chemical factories are quite unique. Electronics must be explosive proof and government certified. Workers in Japan are aging, and younger workers are not interested in chemical factory jobs. This leads to a bigger challenge--- Mitsubishi Chemical’s body of knowledge (workers) is retiring and there are no apprentices.

In conclusion, from everything I learned, Japan needs robots. If nothing else, to capture the body of knowledge from older workers and translate that into "hand crafted automation", in the form of robots. Why I say "hand crafted automation" is due to the fact many human processes in chemical factories are unique and require specialized, i.e. "hand crafted" robots to achieve the desired task. I witnessed so many opportunities for automation across all the factories I visited, and I have a good idea on how to address them. But automation involves risk and strategic vision—balancing practical ideas with automation uncertainties. Japan is a risk adverse country but it also strives for new robot technologies and automation. The tension between these two realities is palpable. We experienced this tension during our project, my colleagues were keen for us to build more robots but they were severely handicapped by limited budgets and restrictions from upper management. The tension motivated SRI and the team which rose to the challenge with flying colors. Even with these restrictions placed upon us, we achieved our goals and ushered a new era of automation for Mitsubishi Chemical, and ultimately Japan.

Bob Christopher I Robot Advisor

?