ERNI is a software development company specialized in medical diagnostics, and they have been working on the issue of implementing digital twins to predict when a technological product will fail.
We talked to Didac Lopez, head of Software Quality at ERNI Spain, so that he could explain the matter better.
What is the difference between a digital twin and a simulation?
There is a lot of simulation within a digital twin and sometimes they are concepts that are confused.
Speaking of software development, a simulation reproduces a series of operations, more or less configurable, against an interface or in response to an action of the system under development. It helps us develop different components in isolation.
On the other hand, a digital twin is an entity of its own that represents a process, system or object, doing so in real time and with interaction with data and external physical systems. This means that the behavior of this digital twin is modeled with data from the real twin or business logic.
Let’s see it with an example:
We have a web service already in production that is gaining popularity. We want to know how the system will behave in a high demand scenario.
On the one hand, we can simulate this high demand through performance tests. The most critical operation at the business level is chosen and thousands of concurrent requests are launched. With this we can optimize this part of the system.
On the other hand, we can create a digital twin of the audience using the data we have about the behavior of users within the service. The Digital Twin will be used to recreate this high demand scenario. Helping us to optimize the system as a whole based on real experience so far.
How can a digital twin predict when a technological product will fail?
Predictive maintenance allows us to anticipate failures in a product through statistical calculation. We rely on data collected over the lifetime of this product and others like it.
This is a characteristic of “Machine Learning”. Product data operating in different contexts are collected, a mathematical model is generated that learns from them and makes predictions about future performance based on the current performance of a specific product.
So where does the digital twin come in?
In order for a Digital Twin to be able to carry out predictive maintenance and warn of a probable failure, it must incorporate the mathematical model and evaluate it with the current data that comes from its real twin.
The digital twin also continues to feed the model with this data so that it can learn and adapt. An advantage of this is that the digital twin, knowing the business logic, can add quality to the data used.
What percentage of reliability do you work with?
The reliability percentage will depend directly on the quality of the data that can be collected, the amount of data available and the algorithm chosen to implement the mathematical model.
In any case, it is important to note that it is not always correct to work with a reliability percentage. Let’s see it with an example:
- Let’s say we have a printer that breaks the head 1 time out of 100 prints.
- Let’s say that our model always results in the impression that the head is in good condition
- This model works with 99% reliability, but the reality is that it is not useful.
There are other metrics that may be more interesting depending on the specific case that the model faces and the consequences that a false negative and/or a false positive have for the business.
What other benefits do working with digital twins have?
Beyond predictive maintenance, we are talking about predictive service. This means that we are no longer just looking for a possible system failure. But to help predict other aspects of the business that also add value. For example, predict what the water consumption of a certain campus will be, enabling those responsible for its management to make logistics decisions.
Within the world of IoT development, access to the physical device or “Hardware” that is going to execute our application is required. This Hardware in turn is usually physically interconnected with other components (antennas, sensors, gyroscopes,…) which in turn are interacting with the world around them.
Working directly with the physical environment is costly and simulating a specific operation is not very flexible. A digital twin of the physical environment allows us to develop in a virtual world and adjust the configuration of components in an agile way. In addition, we have observed that this type of strategy allows rapid scaling of development and testing teams, especially in those scenarios where the device used is expensive in time and money.
Another use of digital twins is prototyping. We helped one of our clients with a laboratory design project. In this case there was a “digital twin” for each of the laboratory elements (machines, personnel, …), they were placed in the virtual space and tests were carried out to assess the most efficient distribution before moving all the equipment. lab’s material.
In complex systems where there is a central component in charge of monitoring, orchestrating or managing multiple devices that are heterogeneous and dynamic. We get a significant advantage by having digital twins of these to generate test scenarios that would otherwise be very difficult to achieve
In conclusion, Predictive Maintenance is one of many uses for a digital twin. The inclusion of this technology can help in prototyping, product development, development of associated products, control and monitoring in the future. By applying the digital twin in these phases, we shorten the time needed to bring to market and evolve a new technology in a physical device that is not yet capable of mass production or whose cost is very high.
What kind of products does this solution work with?
Any product is likely to benefit in one way or another from the concept of “digital twin”. As we have seen, from web applications to IoT development.
Although where it has a higher value is for those products that are difficult to study throughout their useful life such as vehicles, industrial machinery, medical machinery, etc.
We have experiences applying digital twins in smart devices for autonomous vehicles, next-generation 3D printers, robotics, and medical diagnostic devices.
How is the owner or manufacturing company notified that a product is about to fail?
Back to the point of predictive service. What really adds value to our clients is that all these predictions and information from the “digital twin” are accessible, visible and clear. The objective of this service is that the data collected enables business decision-making.
This type of predictive solutions is usually accompanied by an application capable of managing and configuring these digital twins, of representing the information and the alerts are perfectly automated.
Is less electronic waste generated thanks to digital twins?
Without a doubt, in different phases:
In the prototyping phase Having a digital representation of your product that allows you to work until you find the optimal design saves electronic material. In the case of the electronic boards that many items of our day to day have inside, there is a minimum “batch” that must be manufactured so that the manufacturing cost is profitable. With each design change the entire previous “batch” must be thrown away and cannot be reused for anything else.
In the testing phase. Having the possibility of exposing your product to multiple test scenarios in the virtual world means that not only electronic waste but also other associated expenses (such as fuel) are mitigated. For example, some car components are tested by installing them and driving the car for thousands of miles.
In the monitoring phase, better maintenance of the devices extends their life.
Thanks to ERNI for their time and for sharing their knowledge.
