Big Data Clustering for Automotive Safety
News is plentiful, especially in the era of social media. You can always ignore fake news or take time to validate it by finding out common words in the header or within its content. Over time, your search returns results from a set of rules used for grouping certain words that may help you to decide if the news is reliable. This classification or grouping is an example of the application of clustering in Big Data.
Let’s switch gear now to how clustering is being applied to enable the era of autonomous driving. In the case of automotive safety integrity level (ASIL), the grouping of safety levels for different automotive components will be critical. ASIL determination supports the initial phase of many automotive system developments.
ASIL evaluation results are critical for vehicle safety, with all potential hazard and danger scenarios evaluated for specific automotive components. For example, unexpected airbag inflation or inadvertent braking by the radar cruise control system should be assessed and managed in advance.
Automotive makers take this assessment process very seriously, starting with the most stringent simulation tests on integrated circuits (IC) before any component is assembled into the car. Using an oscilloscope, automotive engineers can view and analyze the different outputs by injecting a series of stimuli. A huge number of waveform data is collected over time, and the need to review them manually is inefficient and prone to human errors. Furthermore, trying to group or cluster them based on parameters or shapes similarity manually would be almost impossible.
Keysight created a software solution, PathWave Waveform Analytics (PWA) to help users efficiently and effectively identify anomalies among the huge amounts of waveform data. Clustering is one of its machine learning capabilities, where similar waveform shapes will be grouped together.
Once the master channel is selected, it determines the same time segment for all channels. Then, with the number of clusters selected, the user can view the cluster displays in either Quick or Detail mode.
The Quick mode allows fast overview of waveform clusters from its database. It uses pre-sorted tag information, but its accuracy is limited by the tagging similarity threshold which is how similar the classification of the spikes will be, with the value ranging from 0 to 1 input during data ingestion. With the usage of pre-sorted tag information, the typical computing time is very fast. On the other hand, the Detail mode offers more precise analysis capability as it uses lossless database information. Further details can be seen by performing sub-clustering, where the data is drilled down to the next level with finer breakdown of waveform groups from the selected cluster.
The recommendation is to use Quick mode first, then you can decide if you want to use Detail mode. If Detail mode consumes too much memory, the number of members in each cluster will not fit into the memory. This is where the Retag mode offers another option for viewing the waveform clusters.
From day to day application of grouping different information into reliable or fake news, to the critical differentiation and grouping of ASIL components with the help of PathWave Waveform Analytics, we have seen the powerful application of clustering.
Perhaps the next slew of validated news results from clustering we look forward to receiving is, ‘World Health Organization declares the world is no longer in pandemic’.
Please visit below websites for more information.