Understanding Diagnostics, Maintenance, and Predictive Analytics

In the previous chapter, we implemented I-IoT platforms based on open source, Amazon Web Services (AWS), Original Equipment Manufacturer (OEM), Google Cloud Platform (GCP), and Azure. We also introduced some important topics, including storage, time-series, microservices, protocols, and data processing. Of these, the latter is without a doubt the most valuable topic. General Electric (GE) has monitored aircraft engines and power generation systems for 15 years and claims that just a 1% improvement in efficiency can add $276 billion in profit to major industries. To support these claims, we need to learn from our data and build a new generation of analytics. In the previous chapter, we implemented a few simple analytics. In this chapter, we will discover the most important use cases for I-IoT analytics.

In this chapter...

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 the complete solutions for the exercises in this chapter, go to the repository for this book on GitHub: https://github.com/PacktPublishing/Hands-On-Industrial-Internet-of-Things.

Jupyter

If you have installed Python you may want to use a development IDE. One of the best IDE is the Jupyter Notebook. To install Jupyter, run the following commands:

$ pip install jupyter

To run the notebook, run the following command from the command-line:

$ jupyter notebook

If you want to avoid installing Python and Anaconda, you...

I-IoT analytics

Analytics are one of the most important aspects of the I-IoT and without a doubt are the most complex. To manage complex systems, we need advanced techniques. Because these technologies are very advanced, they can provide valuable and reliable insights, reducing false positives and false negatives. We can think, for example, of the analytics applied to avionic systems or those that apply in safety contexts; clearly, we need high reliability.

Use cases

Over the last five years, the authors of this book have collaborated to develop more than 150 different analytics and thousands of simple rules that can be applied to thousands of different assets. The following are the most important use cases of these analytics...

The different classes of analytics

Analytics can be applied for different purposes and using different techniques. They are generally separated into three different families: descriptive, predictive, and prognostic. These are ordered from the simplest to the most advanced.

Descriptive analytics

Descriptive analytics are the most basic class of analytics. They analyze past data to try and provide a broad view of the fleet. In other words, descriptive analytics tries to answer the question What happened? These analytics use data mining, aggregation, or visual intelligence techniques to understand the status of the assets. In the I-IoT, the most common technique used for descriptive analytics is KPI monitoring.

...

I-IoT analytics technologies

From a technical point of view, I-IoT analytics uses multiple modules—machine learning (ML), anomaly detection, physics kernels, risk analysis, feature engineering, signal processing, optimization methods, simulation methods, damage prognostics, data quality and imputation, and surrogate-modelling. We can identify two way to build analytics, either based on practical rules or a mathematical model.

Rule-based

Rule-based analytics use knowledge about a variable or a particular feature to build a decision-based algorithm. Rule-based analytics can either use expert systems, classifiers, or rule-based ML. Rules-based analytics, for instance, can translate human knowledge or empirical rules into...

Building I-IoT analytics

Identifying the right parameters and techniques is the most important part of building I-IoT analytics. In 1977, John Tukey wrote the book Exploratory Data Analysis. The following method is based on EDA and is applied to the new cloud DevOps requirements. The following are the steps that we need to accomplish:

Analytics development workflow

Step 0 – problem statement

The first step is to always define the scope of the problem, the constraint of the problem, and the expected behavior. Then, we should align the business expectation with the technical expectation. It is a good practice during this step to define the success of our analytics. For example, you might accept 50% false positive results...

Understanding the role of the infrastructure

In this book, in Developing our batch analytics with Airflow section in Chapter 8, Implementing a Custom Industrial IoT Platform, in AWS IoT Analytics section Chapter 10, Implementing a Cloud Industrial IoT Solution with AWS, in Dataflow section in Chapter 11, Implementing a Cloud Industrial IoT Solution with Google Cloud, and in Azure analytics section in Chapter 12, Performing a Practical Industrial IoT Solution with Azure, we discovered how to build analytics based on the cloud or, generally speaking, a centralized infrastructure. We also learned about cold paths and hot paths. From a theoretical point of view, the implementation of analytics should be agnostic with regard to where they are deployed and how we push data. Unfortunately, this is not always the case.

Analytics are strictly coupled with the support they want from the...

Deploying analytics

Although analytics should be agnostic with regard to how the data is fed to the platform, we have to consider several potential pitfalls that can affect the efficiency of the analytics. There are several strategies that we can use to feed I-IoT data to the platform:

• Bulk ingestion, for example, one file daily
• ­Small portion, for example, one file every five minutes
• Data streams, where files are fed continuously with a small latency

Data is also affected by several issues:

• ­ It might be in the wrong order. For example, a data point at 18:00 might be sent at 18:10 and a data point at 17:59 might be sent at 18:11.
• ­It might be of a bad quality.
• It might have holes in it.
• It might have anomalous spikes in it.
• ­It might be frozen. This refers to a situation where you have a suspiciously flat number for a long time.

These issues are illustrated...

Open System Architecture (OSA)

The OSA summarizes the concepts we discussed previously in a single picture. In the Prescriptive analytics section, we highlighted how CBM is one of the most valuable aspects of I-IoT analytics and how it can support insights from descriptive analytics all the way through to prescriptive analytics. The OSA for CBM is depicted as follows:

The OSA model applied to the I-IoT

The OSA for CBM is the most popular framework. It consists of six layers—Data Acquisition, Signal Processing, Condition Monitoring, Health Assessment, Prognostics, and Decision Support. All of these steps can be implemented in a cloud-based architecture and partially on-premises, as explained in the previous sections. In particular, signal processing and condition monitoring can be implemented by using on-stream analytics or, generally speaking, rule-based engines. The health...

Analytics in practice

Let's now put what we have learned into practice. We are going to build a diagnostic analytic and a predictive analytic. We will develop an anomaly detection algorithm for an airplane and a predictive algorithm for an oil and gas refinery. We want to remain as generic as possible, so we won't make any assumptions about the system that we are going to monitor.

We will develop these two use cases with Python, SciPy, NumPy, Seaborn, and Pandas. We will assume that Anaconda 5.2 or Python 3.7 are already installed on your system.

For your convenience Jupyter Notebook are available at the official Github repository https://github.com/PacktPublishing/Hands-On-Industrial-Internet-of-Things . To work with Jupyter Notebook, open command console on the Chapter13 directory, then you can run from command console:

jupyter notebook

...

Summary

In this chapter, we have explored the most important classes of analytics in the I-IoT from a theoretical point of view. We have looked at the most important use cases, including CBM, diagnostic analytics, prognostics, and predictive analytics. We also discussed the relationship of the analytics with the data in terms of model accuracy and data processing. Finally, we implemented a diagnostic algorithm, the anomaly detection exercise, and a predictive analytic model. Anomaly detection and production prediction are two of the most common I-IoT algorithms.

In the next chapter, we will focus on the kernel of the analytics—the digital twin.

Questions

1. Which of the following technologies is most applicable for cloud stream analytics?
   1. Machine learning
   2. Stateful
   3. Simple threshold
   4. High sampling analysis
2. Which of these statements best describes the concept of prognostics?
   1. The time before a component breaks
   2. The probability that a component will break
   3. The remaining life of a component
   4. The prediction of the remaining life of a component
3. What's the EDA?
   1. Explorative Discovery Architecture
   2. Explorative Data Analysis
   3. Explorative Diagnostic Application
4. What's the OSA?
   1. A cloud architecture
   2. An open software for CBM
   3. A framework for diagnostics
   4. An open framework in CBM systems

Further reading

Additional resource can be found at the following links and books:

• AWS Free RTOS: https://aws.amazon.com/freertos/
• General Electric Digital Twins: https://www.ge.com/digital/predix/digital-twin
• MIMOSA: http://www.mimosa.org/mimosa-osa-cbm
• Anomaly detection Algorithm based on Predix: https://www.predix.io/analytics/analytic.html?id=2004
• Python Anomaly Detection at Oracle: https://www.datascience.com/blog/python-anomaly-detection
• Prescriptive versus Predictive Analytics: https://www.datasciencecentral.com/profiles/blogs/prescriptive-versus-predictive-analytics-a-distinction-without-a