Case Study – How Seagate accelerates anomaly detection at Scale with Vanti
The automotive industry is no stranger to strict safety and compliance guidelines. When it comes to LiDAR components, even the smallest gaps in quality assurance can lead to disastrous results in the real-world technologies that rely on these components.
With demand for LiDAR technology at an all-time high, Innoviz needed a scalable solution that allowed them to fulfill the increasing demand for their new MEMS sub-module without compromising on quality.
Discover how Innoviz managed to increase their throughput by 9% during the New Product Introduction (NPI) phase using Vanti Analytics proprietary Early Fault Prediction technology.
Innoviz is a leading manufacturer of high-performance, solid-state LiDAR sensors for automotive, drone, logistics, sidewalk delivery, and many other industries. The company is the first LiDAR provider to be selected by BMW for the mass-production of Level 3-5 autonomous vehicles and has partnerships with the world’s leading Tier 1 automotive suppliers HARMAN, HiRain, Magna, and Aptiv.
NPI is typically seen as an unstable stage in the product life cycle. Innoviz’s product manufacturing process includes the assembly of MEMS and complex optomechanical sub- modules that require high precision and frequent manual calibration of optical elements.
This made it increasingly challenging for the company to meet production goals while still ensuring product quality, continuously optimizing their production line, and performing manual and semi-manual calibrations.
Boosting throughput, reducing delivery times, and improving failure detection rates were top priorities for Innoviz. While the company did have a system in place to detect faulty units, these units were being detected far too late in the process, leading to a large number of defective units being shipped from Europe to Israel as production increased. This resulted in wasted time, money, and effort.
The company also needed to optimize usage for the materials and components required by their customized and expensive semiconductor modules in the manufacturing process.
All of these initiatives resulted in an ambitious goal to predict 70% of faulty units with a minimum accuracy of 90%.
Boosting throughput, reducing delivery times, and improving failure detection rates were top priorities for Innoviz.
Innoviz knew it needed to partner with a proven technology company specializing in data analytics within the manufacturing sector. That’s why they decided to partner with Vanti Analytics.
Vanti Analytic’s Early Fault Prediction technology gave Innoviz a powerful model that could provide real-time predictions for units that were likely to fail at later stages in the quality assurance process.
However, prediction alone wasn’t enough. The model also needed to provide the Innoviz team with immediate fault root cause so detected issues could be addressed and solved on the spot without requiring specialized data science skills.
Vanti used the submodel unit’s historical data over a one-month period to train and get the model production-ready in just two hours. The model was autonomously trained using Vanti’s proprietary machine learning IP. It was then deployed to monitor and detect faulty sub-module units that passed a calibration inspection, ensuring only high-quality units made it to the final stages of assembly.
Detailed root cause reports are now provided for each unit that is predicted to fail, allowing the Innoviz team to identify and resolve issues on the go. This was made possible by specifying parameter clusters that had a high probability of triggering a failure. Examples of these parameters include physical measurements, voltages, and temperatures.
Vanti used the submode unit’s historical data over a one-month period to train and get the model production-ready in just two hours.
Innoviz now has a solution in place that allows them to drive the most value out of their manufacturing data. The company has managed to successfully predict 70% of faulty units, increasing their overall throughput by 9% in the process.
The ability to proactively identify faulty units has reduced the number of faulty units that make it to the later stages of the quality assurance process. This has led to significant cost savings by reducing the amount of manpower required to detect and troubleshoot faulty units throughout the manufacturing process.
Additional insights gained through the advanced analysis and reports allow them to better understand what the prediction model can do, how it makes decisions, and how it impacts other areas of the business. This has led to improved decision-making across the organization.
The early successes achieved by deploying the Early Fault Prediction model have led to increased confidence in the data they’ve collected. This initial deployment has encouraged the company to further explore how predictive analytics can optimize other bottlenecks in the manufacturing process.
The early successes achieved by deploying the Early Fault Prediction model have led to increased confidence in the data they’ve collected.