Machine Learning for Industrial IoT Applications

Advanced Prediction of how long a part will work before performance degrades or stops

Many industrial applications include, but not limited to:

• 5G cell phone towers

• semi-conductor companies

• manufacturing

• mining

• aerospace

• large/heavy equipment

• farming

• autonomous vehicles

• military drones

• commercial airplanes

• oil rigs

• data centers

By combining machine-to-machine communication with industrial big data analytics, IIoT is driving unprecedented levels of efficiency, productivity, and performance. And as a result, industrial companies in original equipment manufacturing, chemicals, food and beverage, automotive, steel, and many other industries are experiencing transformative operational and financial benefits.

For industrial organizations understanding how to to digitize industrial assets, systems, and processes to better understand, predict, and optimize industrial performance to achieve improved business outcomes.

Financial impact to industry

• GE reports that customers can increase reliability from 93% to 99.49% in less than two years, reduce reactive maintenance by 40% in less than one year—contributing to significant maintenance cost savings, and save $360K by predicting a power outage in a gas plant.

• According to Gartner, ““By 2021, half of large industrial companies will use ML on IoT, resulting in those organizations gaining a greater than 10% improvement in effectives.”

• An IoT sensor can represent a functional asset (including fixed or rotating equipment), a component (a pump, motor, etc.), a system of assets (a turbine, jet engine, MRI machine, etc.), or a fleet of assets (a wind farm, a combined cycle unit, an aircraft, etc.). The level of representation is determined by the targeted business outcomes, e.g., reducing unplanned downtime of a compressor compared to optimizing power generation output, to preiciting remaining useful life can save millions to billions annually.

Data Provided by the really, really smart folks at NASA for this example:

Most data that comes from IoT sensors is not time series based, and can oly be time-series by the way you store data. This solution doesnt require that and built-in functionality handles time varying data well.

This is a rapid way to predict the Remaining Useful Life (RUL) of an engine using an initial dataframe of time-series data from NASA [Turbofan Engine Degradation Simulation Data Set](https://ti.arc.nasa.gov/tech/dash/groups/

To load the data, you need to download the data yourself. Download and unzip the file fromhttps://ti.arc.nasa.gov/c/13/ and place the files in the 'data' directory

Highlights

• Automated feature engineering forIoT devices
• Auto formats sensor data into time-series data
• Advanced Prediction of Remaining Useful Life (RUL), custom primitives, and hyper-parameter tuning

Running the tutorial

1. Clone the repo

2. Install the requirements

   pip install -r requirements.txt
   

3. Download the data

   The data is from the NASA Turbofan Engine Degradation Simulation Data Set and is available here

   To run the notebooks, place the following files in the 'data' directory: train_FD004.txt, test_FD004.txt, RUL_FD004.txt

4. Run the Tutorials notebooks:
   

   jupyter notebook
   

   The utils.py script contains a number of useful helper functions.