The Simplicity of Plug-And-Play: Leveraging Machine Learning Using BrickML

Machine learning. It’s a common phrase that has been floating around automated industrial spaces for several years, finding traction in a few specific avenues of engineering. Unfortunately, machine learning systems often require enterprise-level systems along with expensive software, and they are not as easily understood by technical professionals with a more hardware-centric background.

 

The new BrickML for machine learning at the edge of production. Image used courtesy of RELOC

 

How Does Machine Learning Work?

At a high level, machine learning begins with collecting data from a device, equipment, or process. This data is then fed into algorithms and AI to help predict when equipment might start to break down, or when quality issues might start to creep up. The software learns how the machine or process interacts with its own environment. It might be said, then, that machine learning is a subset of AI, but with a very specific purpose to those AI algorithms.

By analyzing interactions and performance metrics, the software can accurately predict a number of issues before they occur, allowing factories to save money and reduce wasted products.

Recently, RELOC, a manufacturer of industrial wireless components, IIoT devices, and edge computing systems, has teamed up with Edge Impulse, a machine learning framework, and released a compact edge machine learning device that promises a plug-and-play method to machine learning.

 

BrickML by RELOC

BrickML box is a stand-alone device that can be mounted directly onto operational equipment and collect data regarding vibrations, temperature, humidity, noise, and voltage. These readings are then funneled into specialty-designed algorithms.

The BrickML, which is made as a joint effort between RELOC and Edge Impulse, is fully integrated into Edge’s industry-leading machine learning software platform. The BrickML comes equipped with a 32-bit ARM cortex built with a 200 MHz clock speed and up to 2 MB flash memory. Each variation of the BrickML has Bluetooth 5.1 and high-speed USB. There is also an enhanced version with CAN-FD and Ethernet connectivity.

The unit can be powered with 5 VDC from a USB or 12-24 VDC for some variations. The rugged exterior can withstand temperatures of -40 °C to 85 °C. The data is collected from four sensors a Knowles microphone for audio, Bosch 9-DOF IMU for vibration detection, voltage inputs for ADC and MCSA applications, and a Renesas HS3001 temperature and humidity sensor.

The BrickML is only 40 mm x 23 mm x 5 mm, making this device small enough to place on virtually any medium-sized device.

 

Predictive maintenance is a bigger problem for aging equipment, but its applications can be easily expanded to new installations. Image used courtesy of Unsplash

 

Edge Impulse Software

The software obviously has a major impact on machine learning devices, since this is where all the calculations are performed and where all the data is stored. Edge Impulse is a developer platform that can be deployed on virtually any computer or micro-computer. The software can then be configured to collect data from multiple sources and analyze that data with specific models.

Edge Impulse is used on applications ranging from wearable devices to industrial devices, and it can be integrated into existing workflows with native Python integrations or Node.js APIs. There are even enterprise-level admin APIs available.

 

The software is a massive component of any good preventive/predictive maintenance program. Image courtesy of Edge Impulse

 

Applications of the BrickML

Equipment downtime can cost companies a lot of money, compounded by every passing second, so it comes as no surprise that companies are investing heavily in machine learning applications that can learn the process and predict when equipment will need to be maintained. In process automation, when the quality of your product starts to drift, the wasted product becomes very costly and can affect the environment. With advancements in machine learning algorithms, we can predict exactly when the process might start to see a drop in quality.

The BrickML is a very small device, designed to monitor specific operating factors on equipment. One of the most apparent usages is an application to motors. An electric motor has moving parts that wear down and will, at some point, require servicing. If that motor is spinning non-stop, with a very constant voltage and load, you might be able to fairly easily predict when it will need servicing. But motors are not usually operated like that in the field. They are often overloaded, subjected to frequent stops and starts, and used in an operating environment that takes a serious toll on the lifespan of the motor.

By using a machine learning device like BrickML, you can begin harvesting long-term data on the motor, providing a clearer idea of when it might need to be serviced, saving the company downtime and loss of product in the future, all in trade for the present investment into innovative device and software applications.