In AI, Data Science, Digital Manufacturing, How-To Articles

In manufacturing, obtaining insights from production data can deliver transformational improvements in efficiency and quality. In fact, IDC predicts that by 2021, 20% of the world’s largest organizations will operate based on data and analytics derived from artificial intelligence (AI), machine learning, and the Internet of Things (IoT).[1]

However, there are many obstacles to prevent you from realizing this value. The fact is that manufacturing data is, well, messy. AI-based technologies can automate the processes of preparing this data to deliver business value, and, according to the IDC report, speed production by up to 25%.

From One-Time to Real-Time

Let’s take a simple example: Your facility gathers information from a temperature sensor that takes regular readings in a paint booth. It is reliable at first, but the data starts to drift as the sensor wears out or breaks down. Over time, it even begins surging and slacking, or showing unusual (and possibly false) values. Worse, these spiking numbers over time may become the norm. Analysts can’t be sure whether the lamps are working correctly, how long it is taking to dry the components, or whether surfaces are at risk due to improper heat. The result is unscrubbed, noisy data that is difficult to use in measuring KPIs or root cause analysis.

Managing noise from a single sensor for a single analysis is an annoying inefficiency.  Attempting to manage potentially noisy data from many sensors, over time, for a wide variety of applications, is a severe impediment to creating operationalized analytics. Instead, data scientists need a solution to sustainably manage a variety of raw data challenges, allowing them to focus on the analytics instead of data wrangling.

Automatically Cleaned Data Streams

 To address this, tools such as Sight Machine help you create real-time streams of clean, ready-to-use data. Such a resource includes:

  1. Processing data into flexible data models, including rules for detecting and managing outliers, drifting values, and other draft data
  2. A data stream that is continuously, automatically cleaned
  3. Easy-to-use interfaces to receive the pre-processed data
  4. Outputs presented in a way that explain the results effectively to a variety of operators, regardless of their experience

In this environment, you can more quickly address a drifting data source, such as our temperature sensor, by retraining the AI with a continuous stream of readings. An ongoing flow of information makes it easier to adapt to and accommodate data fluctuations as part of the process.

Clear and Interpretable Results

Even if the data is clean and readily usable by data scientists, subsequent analysis may not be usable by the ultimate end users on the production floor.  Manufacturers are understandably conservative about trusting data in the midst of a changing, challenging process, and operators need clear and interpretable results.  This enables them to make immediate decisions on the production line – for example, locating and replacing that faulty temperature sensor.

To accomplish this, good visualizations of information are essential, based on real-time streams or on historical data “snapshots.” However, the most useful visualizations are rarely the three-dimensional rotating graphs or tables (with lots of Greek letters) that make for great marketing. In fact, the results of several of our projects, even those based on complex chains of algorithms, can be visualized with easy-to-read Pareto charts.

Managing the Future of Data

By collecting, analyzing, and interpreting the very large data sets produced by factory and channel systems, manufacturers can take advantage of amounts of information previously too great for humans to evaluate. At the same time, the single greatest challenge today is the skyrocketing amount of useless, noisy data. A successful AI project is one that converts this noise into valuable information, becoming a critical part of doing business as usual.

Please contact us for a live demonstration of how Sight Machine’s AI-driven platform and CI services can achieve these goals.

[1] “FutureScape: Worldwide Manufacturing Predictions 2018,” IDC Report

Why Sight Machine Uniquely Answers Your Manufacturing Needs Better manufacturing analytics across the enterprise The only AI-enabled platform for digital manufacturing built from deep manufacturing experience, understanding, and a history of results Learn More

Recommended Posts

Leave a Comment