Ted Hisokawa
August 31, 2024 00:55
NVIDIA’s RAPIDS AI powers predictive maintenance in manufacturing with advanced data analytics, reducing downtime and operating costs.
According to the International Society of Automation (ISA), 5% of factory production is lost each year due to downtime. This equates to a global loss of approximately $647 billion for manufacturers across a wide range of industries. According to the NVIDIA Technical Blog, the most important challenge is to minimize downtime, reduce operating costs, and predict maintenance requirements to optimize maintenance schedules.
LatentView Analysis
A key player in this space, LatentView Analytics supports multiple Desktop as a Service (DaaS) customers. The $3 billion DaaS industry is growing 12% annually and faces unique challenges in predictive maintenance. LatentView has developed PULSE, an advanced predictive maintenance solution that leverages IoT-enabled assets and cutting-edge analytics to provide real-time insights and significantly reduce unplanned downtime and maintenance costs.
Remaining Use Life Use Cases
A leading computing device manufacturer wanted to implement effective preventive maintenance to address component failures occurring in millions of leased devices. LatentView’s predictive maintenance model aimed to reduce customer churn and increase profitability by predicting the remaining useful life (RUL) of each machine. The model aggregates data from key thermal, battery, fan, disk, and CPU sensors and applies it to a predictive model to predict machine failures and recommend timely repair or replacement.
The challenges we face
LatentView faced several challenges in its initial proof of concept, including computational bottlenecks and extended processing times due to massive data. Other challenges included handling large real-time data sets, sparse and noisy sensor data, complex multivariate relationships, and high infrastructure costs. These challenges required tools and library integrations that could dynamically scale and optimize total cost of ownership (TCO).
Accelerated predictive maintenance solution with RAPIDS
To overcome these challenges, LatentView has integrated NVIDIA RAPIDS into the PULSE platform. RAPIDS provides an accelerated data pipeline, operates on a platform familiar to data scientists, and efficiently processes sparse and noisy sensor data. This integration has resulted in significant performance improvements, enabling faster data loading, preprocessing, and model training.
Create faster data pipelines
Leveraging GPU acceleration parallelizes workloads, reducing the burden on CPU infrastructure, resulting in lower costs and improved performance.
Working on known platforms
RAPIDS leverages packages that are syntactically similar to popular Python libraries like pandas and scikit-learn, enabling data scientists to accelerate development without requiring new skills.
Exploring dynamic operating conditions
GPU acceleration allows models to seamlessly adapt to dynamic conditions and additional training data, ensuring robustness and responsiveness to changing patterns.
Processing sparse and noisy sensor data
RAPIDS significantly speeds up data preprocessing and effectively handles missing values, noise, and irregularities during data collection, laying the foundation for accurate prediction models.
Faster data loading and preprocessing, model training
Apache Arrow-based RAPIDS capabilities speed up data manipulation operations by more than 10x, reduce model iteration times, and enable multiple model evaluations in a short period of time.
CPU and RAPIDS performance comparison
LatentView performed a proof of concept to benchmark the performance of RAPIDS on GPUs and CPU-only models. The comparison highlighted significant speedups in data preparation, feature engineering, and group-by-group operations, achieving up to 639x improvement on certain tasks.
conclusion
The successful integration of RAPIDS into the PULSE platform has resulted in compelling predictive maintenance outcomes for LatentView’s customers. The solution is currently in the proof-of-concept phase and is expected to be fully deployed by Q4 2024. LatentView plans to continue leveraging RAPIDS to model projects across its manufacturing portfolio.
Image source: Shutterstock