AI-powered image analytics for identifying oil degradation and predicting RUL of vacuum pumps​

Largest vacuum pumps manufacturer leverages AI-driven image analytics for oil condition monitoring, enabling predictive maintenance, reduced unplanned downtime and warranty costs for vacuum dry pumps operating in critical semiconductor fabrication process.

Learn how. Talk to our Specialist

About client

Our client is a largest global manufacturer of industrial vacuum pumps used across semiconductor, pharmaceutical, chemical, and advanced manufacturing industries. These pumps are mission-critical assets where performance, uptime, and maintenance efficiency directly impact critical semiconductor production outcomes.

A key factor influencing pump performance is oil condition. Oil degradation affects reliability, efficiency, and service intervals. Traditionally, oil health assessment relied on manual visual inspection through pump sight glasses during maintenance cycles.

However, this manual approach was subjective, inconsistent, and limited in its ability to detect early degradation patterns or support predictive maintenance strategies.

The Challenge: Why manual oil monitoring was no longer sustainable

Oil condition monitoring is critical for ensuring pump reliability and avoiding costly downtime. When the process relied on manual inspection and experience-based judgment, several operational challenges emerged:

These limitations prevented the organization from transitioning toward a more intelligent, data-driven maintenance model.

From manual inspection to AI-driven oil health monitoring

Before After
Oil condition was visually inspected through sight glass, relying on technician judgment. Oil images are analyzed using AI models to classify oil health objectively.
Assessment varied across individuals, leading to inconsistent maintenance decisions. Standardized classification ensures consistent and reliable oil health evaluation.
Maintenance schedules were time-based rather than condition-based. Maintenance decisions are driven by actual oil condition and predicted degradation.
No visibility into oil degradation trends or failure risks. AI models provide insights into degradation patterns and Remaining Useful Life (RUL).
No digital system to analyze or store oil condition data. A centralized portal enables image upload, analysis, and result visualization.

Solution: AI-based oil condition monitoring using computer vision

To address these challenges, the client partnered with Saviant to develop a computer vision-based AI solution that digitizes and standardizes oil condition monitoring. The solution is designed as a scalable digital capability aligned with modern industrial product innovation and predictive maintenance goals.

01. AI model for oil health classification

A Convolutional Neural Network (CNN) model was developed using Azure Custom Vision to analyze oil sight glass images.

The model enables automated, objective oil condition assessment, eliminating reliance on manual inspection.

02. Remaining Useful Life (RUL) estimation

In addition to classification, the solution introduces predictive insights through Remaining Useful Life estimation.

This shifts maintenance from reactive or scheduled to predictive and condition-based.

03. Web-based analytics portal

A web application was developed to make the AI capability accessible to end users. Operators and engineers can:

The portal acts as the foundation for future integration with broader equipment monitoring systems.

Technology foundation

The solution was built using a modern, scalable technology stack aligned with industrial digital transformation needs:

Saviant’s engagement approach: Built to solve the right problem first

The engagement was executed as a Proof of Concept (PoC), emphasizing rapid validation and measurable business value.

  1. Started with problem and value discovery & value design
    Saviant AI consulting & development team worked closely with the client’s engineering and domain experts to understand oil degradation behavior, current inspection limitations, and define where AI could deliver measurable value in improving maintenance decisions.
  2. Identified the right data and approach
    Relevant datasets were curated, cleaned, and labeled by combining historical oil images with operational metadata, ensuring the foundation was aligned to real-world conditions and domain expertise.
  3. Built and refined what mattered first
    Our initial focus was on developing an AI model to accurately classify oil health. The model was iteratively improved through multiple training cycles and continuous validation with subject matter experts.
  4. Validated through real-world usage
    An interactive web portal was developed to allow stakeholders to upload images and test predictions, enabling early validation, feedback, and confidence in the solution.
  5. Planned for scalable productization
    The solution architecture was designed with future scalability in mind, laying the groundwork for transitioning from a Proof of Concept to a production-ready predictive maintenance platform.

The impact so far

By digitizing oil monitoring and introducing AI-driven insights, the client has taken a significant step toward intelligent, condition-based maintenance.

Current adoption and roadmap

Following the successful PoC, the client is progressing toward expanding the solution’s capabilities and industrial adoption. Key focus areas include:

What’s next: Toward full predictive maintenance

The next phase focuses on building a multivariate predictive model that combines multiple data sources:

This will enable:

The roadmap also includes:

Not sure how AI fits into your product or roadmap?
Saviant’s AI Maturity Assessment can help!

Talk to our AI specialist