Implementing a Digital Twin – Advanced Analytics

In the previous chapter, we investigated some of the most interesting classes and use cases to do with I-IoT analytics. We discovered that an analytic can be descriptive, diagnostic, predictive, or prognostic. We also mentioned the remaining useful life (RUL) of an asset or part within the I-IoT.

In this chapter, we will improve our knowledge of I-IoT analytics, using more advanced technologies based on machine learning (ML) and deep learning (DL).

In this section, we will explore the following:

• Digital twins
• Practical examples with DL and ML algorithms
• Platforms on which to build digital twins

Technical requirements

In this chapter, we need the following prerequisites:

• Git: https://git-scm.com/downloads
• Python 3.7: https://www.python.org/downloads/
• Anaconda 5.2: https://www.anaconda.com/download/

To find a complete solution for the exercises, go to the repository at https://github.com/PacktPublishing/Hands-On-Industrial-Internet-of-Things.

Advanced analytics and digital twins

The official definition of a digital twin is a representation of a physical asset. The first experiments with digital twins were developed by NASA to mirror outer space as a virtual image. Later, digital twin technology was used for product design and manufacturing, in techniques such as 3D prototyping. When more complex objects began to produce data, digital twins moved beyond manufacturing into the domains of the I-IoT, artificial intelligence, and data analytics. Having a digital copy of a physical object gives data scientists the ability to optimize efficiency and create other what-if scenarios.

Digital twins are used in the following sectors:

• Renewable energy: To emulate wind turbine plants and optimize production
• Mechanical: For diagnostic purposes, degradation analysis, or what-if scenarios
• Oil, gas, and power generation: To optimize...

Advanced analytics in practice

Let's put into practice what we have learned so far. We will develop a digital twin with Python, Keras (TensorFlow), and Pandas. We assume that Anaconda 5.2 or Python 3.7 is already installed.

Evaluating the RUL of 100 engines

For this exercise, we will use the free Turbofan Engine Degradation Simulation Data Set provided by NASA.

The Turbofan engine dataset is a free database provided by NASA (Saxena and K. Goebel (2008). Turbofan Engine Degradation Simulation Data Set, NASA Ames Prognostics Data Repository (http://ti.arc.nasa.gov/project/prognostic-data-repository), NASA Ames Research Center, Moffett Field, CA).

...

Platforms for digital twins

A digital twins platform is not that much different from the analytics platforms studied in the previous chapters. All it needs is sufficient storage support to save the current status of the model and to manage the life cycle of the digital twin. In other words, digital twins require the model to be tuned initially. They also require the real asset and the digital asset to be synchronized periodically:

The components of digital twins

The platforms that can support digital twins today include AWS, Predix, and Google Cloud Platform (GCP).

AWS

AWS recommends that we use SageMaker as the main platform for ML. With SageMaker, we can define our model to train parameters and hyperparameters. We can also...

Other kinds of I-IoT data

In the I-IoT, we frequently speak about time-series and sensor data. In some circumstances, however, we can work also with unstructured data. This might include images, such as pictures of geological layers, inspection images, or high frequency data, or log files, such as equipment shutdown report files or security report files. Although this data is not technically part of the I-IoT, it should be considered in the data flow.

Summary

In this chapter, we have explored one of the hottest topics of the I-IoT—the digital twin. We learned about the differences between the physics-based and data-driven approaches and we applied these to two use cases—wind turbine and engines. We also looked at some new technologies, such as DL and RL.

In the next chapter, we will apply the exercises that we have carried out in the past two chapters to a cloud-based technology using Azure, GCP, and AWS.

Questions

1. Which of the following technologies is not a neural network technology?
   1. CNN
   2. RNN
   3. LSTNN
   4. LSTM
2. Which of the following is the best definition of digital twins?
   1. A 3D model of a piece of equipment
   2. A digital copy of the design project of a piece of equipment
   3. A digital representation of a piece of equipment
3. What is the difference between a physics-based and a data-driven model?
   1. A physics-based model is based on mathematics, while a data-driven model is based on statistics
   2. A physics-based model is based on design knowledge, while a data-driven model is driven by data
   3. A physics-based model is based on rules, while a data-driven model is based on machine learning or deep learning

Further reading

Additional resource can be found via the following links:

• General Electric Digital Twins: https://www.ge.com/digital/predix/digital-twin
• Predix Digital Twins: https://www.slideshare.net/predixdevelopers/predix-transform-2016-e1-building-the-digital-twin-64632276
• The fall of RNN/LSTM: https://towardsdatascience.com/the-fall-of-rnn-lstm-2d1594c74ce0
• Keras LSTM Tutorial: http://adventuresinmachinelearning.com/keras-lstm-tutorial/
• Defining a Digital Twin Platform: https://medium.com/@iskerrett/defining-a-digital-twin-platform-de67586623de
• IBM Watson Digital Twins: https://www.ibm.com/internet-of-things/spotlight/digital-twin
• Microsoft Digital Twins: https://enterprise.microsoft.com/en-us/trends/microsoft-is-redefining-digital-twins-in-discrete-manufacturing/
• AWS SageMaker: https://aws.amazon.com/sagemaker/
• Google Digital Twin: https://cloud.google.com/blog/products...