Monitoring the health of Ultrasonic Washers
Monitoring the health of machinery using AWS IoT Solutions
This post described how IntelZone implemented a real time monitoring of ultrasonic washers, utilizing IoT solutions from Amazon Web Service as well as our own knowledge of sensor technology.
Optimize the use of Industrial cleaner systems. In September of 2017 the owner of Ultralydvasker.no, an importer of Ultrasonic washers and industrial washers contacted IntelZone to consider the possibilities of installing smart solutions on their machines. After an initial Workshop with IntelZone a problem statement was identified as well as the key components to monitor. The ProSonic Ultrasonic washers, are used frequently and needs a high operational functionality. Before IntelZone was involved, the machinery followed a simple best practice for periodical maintenance. IntelZone was asked to consider the opportunities for monitoring the health of the machinery, to increase time in use and decrease unwanted, unscheduled maintenance. A solution for automatic refilling and reordering of cleaning liquid was also requested. This includes decreased downtime of the machinery, optimized cleaning and automatic refilling and ordering of chemicals.
Install sensors on two existing machines, for data gathering in a test phase of 3 months. The sensors will gather information on vibration, power usage, PH level, temperature, water level and sound. The data collected is then analysed, and using Machine Learning(ML) to look for patterns that may indicate that an unwanted machine incident will take place. After the initial ML model has been developed this will be reviewed after 4 months.
A web based desktop will also be designed. This will work as a real-time monitoring of the status and health of the machinery. It will also be used to send work orders if maintenance is needed. Amazon IoT solutions, such as AWS IoT Core and AWS Greengrass, will be used to connect the sensors and gateway to the Cloud, in a secure way. Data will be stored in a DynamoDB and visualised using Amazon QuickSight. Amazon CloudWatch will be used to monitor the usage of the different AWS solution, and optimize the usage.
To ensure the security and flexibility of the cloud and sensor system, AWS cloud solutions were utilized. This makes sure that communication between the sensor and cloud is secured, while it’s easy to implement on new machines in the future.
Sensors will be installed to control several factors on these key components, such as vibration, power usage, PH level, temperature, water level and sound. The sensors will be controlled using a main gateway, running AWS Greengrass with sensor connected by cable or wireless, where the sensor will be controlled by devices running on AWS IoT Device SDK.
After the sensors are connected and gathering data, the test machines will run for 3 months where all unwanted incidents will manually be registered. After this a ML model will be implemented and improved over the next 4 months, where manual registration of unwanted incidents will be required. The data collected in the test phase will be stored in an S3 files system. After the ML model is ready for production the key data will be stored in an Amazon DynamoDB solution, for later optimization of the ML model. Key data will also be made available in an Dashboard solution running on Amazon QuickSight, for real-time monitoring of the health status of any connected machine.
Two ProSonic ultrasound cleaners, that have been in use for two and four years, were fitted with sensors to monitor, vibration, power usage, PH level, temperature, water level and sound. After the first month in testing some additional vibration sensor was fitted close to key components, after advice from the engineer responsible for maintenance.
After the initial 3-month period the machine learning model was deployed and the first deployment of the health and status desktop was released. Two sound sensors were found to be placed in such a manner that it was hard to use the data collected. One of the sensor was removed, while the other was isolated from unwanted noise.
As of April 20 the latest ML model has been running for 2 months and 75%(3 out of 4) of unwanted incidents was successfully identified, while 50%(2 out of 4) was predicted, by the ML model.