Pick and Place machines needed to be monitored for nozzle performance. An early warning notification system in case of discrepancy, was to be established.
Our team of experts developed Predictive Machine Learning models based on performance data to determine if the nozzle was getting out of control. They also set up a trigger for early warning notifications, which could then be acted upon by the concerned personnel team.
An Edge device from the shop floor publishes the data onto a Kafka topic. The Gateway consumes this data from the Kafka and runs the adaptive business, ML logic to decide the performance of the nozzle. The ML Logic is executed on the Data processing Gateway using a container running on Kubernetes.
If the nozzle performance is found to be out of control, a notification is generated and this data is pushed to the cloud for Analytics. The real-time dashboard renders the performance charts of the nozzle.
Our client wished to to check for solder joint defects during the manufacturing process of their Printed Circuit Board (PCB). By augmenting their SPI defect-level detection, they wanted to improve their panel testing and reduce false positives.
We developed a predictive real-time analysis component for processing. The SPI parametric data flows from the machine into Analytics edge, and into the Data Processing Gateway, which runs Kafka.
The IoT Edge device subscribes to the raw data topic, executes an advanced machine learning model and uses the output to display the risk of testing failure on an HMI device, which the operator can view. Additionally the IoT Edge device will send control commands to the manufacturing line.
A dockerized machine- learning algorithm is deployed on the edge, on premise or in the cloud. The Cloud based ML training continuously improves the capabilities and accuracy of the models. A containerized workload consumes the raw data feed and publishes it to a data lake. The ML Suite enables the turning on and off of the elastic compute infrastructure.
Monitor device sensors to detect failures before they occur
Interpret energy usage patterns and adjust resources in connected networks
Identify unusual patterns in vehicle speed or location
Identify web traffic and potential server degradation
We modeled our solution using the Hierarchical Temporal Memory (HTM) algorithm to automatically learn and model system behavior, and identify irregularities. Such irregularities in machine generated data is then encoded to Sparse Distributed Representations (SDRs) that are in turn, processed by the HTM Learning Algorithms. When an HTM receives input, it will match it to previously learned spatial, temporal patterns and successfully matches new inputs to previously stored sequences.
In this fashion, we created a time series anomaly detection engine capable of being plugged into any incoming streaming source (Kafka) and detecting abnormal behavioral patterns.
The Hierarchical Temporal Memory (HTM) based solution enhances the efficiency of pattern identification.
Establishes a balance in electricity consumption and generation in daily power system operations, guaranteeing adequate power supply even during disturbances like random load.
The Predictive Analysis capability helps prevent outages and shutdowns.
Fast detection of unusual employee activity such as accessing of confidential data, and internal systems. Alert notification system for such unusual employee behaviors.
Identifies abnormal or fraudulent financial trading activities or asset allocations by individual traders.
