Predictive maintenance, implementation in 8 weeks

Many companies have major problems with the widespread implementation of predictive maintenance projects, apart from a few lighthouse projects. Benefit from the 8-week plan for developing a functional pilot.

Hardware developers often obsessively strive for the 100-% solution. In mechanical engineering, this is certainly justified. After all, the machine should ultimately be as robust as possible. In software development, on the other hand, it makes sense to work quickly and agilely. Pilots should be set up quickly, validated with the customer and then sent back into the next development loop. Step-by-step development must be the goal.

Try the following with your Predictive maintenance implementation not to build the egg-laying wool-milk sow. Otherwise, it will be years before your development project is completed. By the time the system is launched, it will already be outdated. Your solution does not have to cover all customer scenarios and be able to predict every problem at the start. It is sufficient to put three bright minds from IT, software and control development together for a week to decide on the rough system design.

A months-long fine-tuning process does not make sense. The result would only be marginally improved. Say goodbye to the idea of a perfect solution. It will have to be adapted in a few years anyway.

We have developed 5 steps for the development of a Predictive maintenance-pilots. We are starting from scratch and the results should be visible after just 8 weeks.

1. team building

To get started, you need a good, interdisciplinary team. The specialist knowledge should be as broadly diversified as possible and cover the areas of machine technology as well as software and control development.

They should not only be specialists in the respective field, but also well-networked and broadly positioned employees. They can bring missing know-how from their contacts into the project and thus lead it to success. We suggest 4 team members for the initial set-up of your mission:

• A project manager - He or she removes obstacles, has the team's back and should have vision and foresight. The project manager can also be an external person. Give us a call if you need help!
• A service technician or machine developer - he knows the typical problems with the machines.
• A control system developer - he knows which data is available where on the machine and can be deducted.
• A software developer - he can quickly build a user interface with tools available on the market and automate data analyses.

Take time from Monday to Wednesday to get the team members on board for your predictive maintenance implementation. Send them to a team-building activity on Thursday and Friday. Preferably to a secluded mountain hut in the Alps or something similar. They should get to know each other.

It goes without saying that the team is given the go-ahead to concentrate on the project during the 8 weeks of 100% and is released from regular activities. Compared to some mammoth projects that extend over several years and often fizzle out without results, the costs for this should be justifiable.

It is also advisable to work outside the company's structures, as otherwise initial coordination with IT can quickly take longer than 8 weeks.

2. build data collector

Take the data from the machine to the cloud! Control and software developers have to select which data from the machines ends up on an accessible server and how. In case of doubt, your creative and solution-oriented team will know better for your specific project, but you may be able to stick to this procedure:

Don't rush into the raw datagenerated by the countless sensors in the machine. The storage and transmission of these gigabytes are too expensive for later scaling.

In addition, they are not very meaningful for your predictive maintenance project. It is better to use the Aggregated machine log file of the system. It is normally used in the central PLC (Programmable logic controller) and stored locally on the machine. That's why we have the control system developer on board.

The aggregated machine log file offers you a number of advantages for your Predictive maintenance implementation:

Now you need to upload this file to the cloud regularly, let's say every hour to start with. Under no circumstances should you take the typical route via the customer network and a VPN tunnel. You would have to drill holes in the customer's firewall, which would require a lot of coordination with the customer. That goes beyond the scope of our 8-week approach.

To get started, simply work with a laptop with a mobile phone card next to the machine and connect it to the machine controller. This retrieves the file from an exchange folder to be set up on the machine's industrial PC and sends it to your address in the cloud.

Use Amazon Web Services (AWS) as the host for your pilot! Many people may break out in a sweat at the thought. But you can switch to other services later or set up your own infrastructure. The advantage is the powerful tools and the fact that you can simply click your application together later with a little expertise. This saves programming work!

3. build Data Analizer

Now you have to deal with the project-related evaluation of the data and find out what can be implemented with the given input. This phase is where your service expert and your software developer can really shine.

Workshops must be held to clarify which messages indicate possible disruptions in the future. This shows the diversity of implementations in the area of Predictive maintenanceas there are virtually no limits to creativity. For example, can sensor messages about pressure drops in your cooling system indicate tubing that is slowly becoming porous?

Are rising temperatures in the control cabinet an indication that the cooler could fail in the future? The team should initially limit itself to 3 to 5 use cases that occur relatively frequently. The software developer can use these analysis algorithms in your Predictive maintenance-application with the AWS toolkit.

This will give you a warning message when potential failures are imminent. Your team should need about 3 weeks for this step. It is the most time-consuming part of your implementation plan and requires a lot of brainstorming.     

If the proposed approach is based too much on the experience of your own team, you can also start directly at this point by implementing a machine learning algorithm that independently searches for anomalies and error correlations. This approach cannot be fully implemented in the planned 8 weeks, but you could lay the foundation for the subsequent development phases in parallel. The company Splunk, for example, offers good AI software for analyzing machine log files. I have already seen good solutions based on this software that have successfully found their way from PowerPoint into reality. - Dr. Simon Tonat

Artificial intelligence is a highly interesting topic. Also in service! Find out more about the use of Machine learning for predictive maintenance.   

4. set up push alarms

The software should now be able to predict possible failures. Set up via your IT Email inbox so that your colleagues in the service department do not have to constantly check the status online. The software developer can ensure that automatic emails are sent to the mailbox when a problem is detected by the algorithms.

You can also consider what additional services you can develop in the future based on the data. Is it possible to draw conclusions about the efficiency of machine use by the customer based on the program sequences? And can consulting services be derived from this? Or is the customer interested in an overview of the utilization of his machine?

There will be some ideas for the next development loop here. This task is mainly the responsibility of your project manager.

The automatic forwarding should work after one week at the latest. The best way to do this is via an interface in your Field Service Management Systemas you have mapped all processes there. Ultimately, however, we are building a Predictive maintenance-pilots and not a complete solution. It may therefore be necessary to make compromises. We will continue to approach the perfect result incrementally.  

5 WOW your customer!

Now comes the best part: call your customer and explain that their machine is about to break down or malfunction. He may initially think this is a scam to increase your spare parts sales.

However, if there is actually an outage a short time later, they will be very surprised. This will ensure that your customer is more open to further data-based services in the future. This is what you should do next:

• - Spread the success story on all channels through PR and marketing. This will also make other customers curious about Predictive maintenance become.
• - Have your project team glorified with a lurid story on the intranet. This will boost their motivation for follow-up projects.
• - Present your success to the management to ensure that they are open to investing in further development.

If you have started with 5-10 customers, the first call from your side should be feasible within a week. Congratulations! You have implemented your first predictive maintenance project in just 8 weeks. An important step for your After-Sales-Department