DC Water: Streamlined Sewer Pipe Inspection Analysis

Cut costs, optimize maintenance assignments and improve customer service.

At a Glance:

  • DC Water distributes drinking water and collects and treats wastewater for more than 672,000 residents and 17.8 million annual visitors in the District of Columbia.

  • DC Water, in collaboration with Wipro, developed Pipe Sleuth to automate the process of identification, annotation, scoring/grading of pipeline health and reporting of pipeline defects. Optimized Pipe Sleuth, with Intel® Xeon® Scalable processors and Intel® Distribution of OpenVINO™ toolkit, resulted in faster time-to-market, cost savings on analysis, and allows for more spending on capital improvements.


Executive Summary
Artificial Intelligence (AI) enables utilities to speed up analysis of sewer pipe inspection videos, while ensuring accuracy. This allows their workforce to focus on making timely maintenance decisions instead of conducting routine inspections. Other benefits include reduced costs, lower rates of error, and fewer disruptions in service.

To help lower the hurdle of deploying AI, utilities can work with solution providers experienced in the challenges that affect this sector. DC Water enlisted the help of Wipro, an IT services provider, and member of the Intel® AI Builders program with expertise in the utility industry as well as computer vision, machine learning, and related technologies.

Sewer Systems: Miles of Challenges
Sewer pipe maintenance is like a form of insurance against future service disruptions. As part of the cost-benefit analysis, utilities are looking for ways to mitigate sewer pipe faults while lowering the cost of inspection.

This is critical given how many miles of sewer pipe lay hidden beneath cities. For example, more than 1,800 miles of sewer pipe crisscross under Washington, D.C.1 Its more than 701,000 residents and nearly 20 million annual visitors depend on the District of Columbia Water and Sewer Authority (DC Water) to collect and treat wastewater.2

The utility’s system dates back to 1810, with pipes built from a variety of materials including brick and concrete, vitrified clay, reinforced concrete, and cast iron.

Infrastructure such as this is typical throughout the United States and other countries. That makes it a full-time job for utilities to ensure their sewer system remains consistent and operational, as wastewater removal is an expectation that customers take for granted.

Standard Inspection Is a Cumbersome Process
Regular inspection of underground sewage pipelines enables utilities to prioritize maintenance tasks that can help prevent leakage, breakage, and blockage. Standard inspection consists of difficult, time-consuming manual activities. In one common method, an operator remotely guides a camera-mounted rover to record video of underground sewer pipes. The operator uses the video recording to produce an inspection log and summary report, flagging anomalies or problems along with their locations.

Quality control staff then reviews the video recordings and reports for accuracy. With thousands of miles of pipe and hours of video to view in real time, it’s a never-ending process. Fatigue, distraction, and differing opinion as to which maintenance items or repairs should be marked a priority can lead to errors.

Moreover, it’s not uncommon for utilities using this method to wait until end-of-day Friday to evaluate a week’s worth of reports before issuing work orders for the following week. That can lead to backlogs. Backlogs can lead to delays. Delays can lead to emergency repairs.

This has motivated utility companies to look for ways that increase the efficiency of the pipeline inspection process.

Automating Sewer Pipe Inspection with AI
DC Water sought a solution that would automate and help to accelerate pipeline video inspection while still maintaining accuracy.

The utility’s management team met with counterparts from Wipro, a company with experience in automation, AI, analytics, and other technologies. Each organization brought its domain knowledge, and together they worked on developing a solution that would eliminate manual review and classification of sewer pipe video scans.

Their goal: optimize infrastructure maintenance in order to save time, improve efficiency, and lower costs. Automating sewer pipe inspection video analysis and reporting could help achieve this.

DC Water and Wipro Collaborated in Their Development of Pipe Sleuth
DC Water applied its deep domain knowledge in qualifying the severity of anomalies to generate comprehensive reports that could seamlessly integrate into its existing workflow.

Wipro combined its strong Digital Signal Processing (DSP) domain skills – in the areas of image and video processing and computer vision – with AI and machine learning technologies in order to solve business problems.

The company’s history includes vast IT system integration experience with both public and private entities. This includes government-owned water providers as well as gas and energy companies in the US and around the world. By providing both technological and sector expertise, Wipro helped guide DC Water on its digital transformation journey, turning the utility’s capabilities and insight into a practical solution.

Applying Computer Vision to Sewer Inspection
The Pipe Sleuth solution automates the process of identification, annotation, scoring/grading of pipeline condition, and reporting of pipeline defects using standards that meet the Pipeline Assessment Certification Program (PACP). These benchmarks were instituted by the National Association of Sewer Service Companies (NASSCO).

“We are using Pipe Sleuth to inspect our sewer network. It is an innovative solution that dramatically increases inspection productivity and significantly reduces costs while at the same time improving the overall defect detection rate. Having the option to run Pipe Sleuth on our existing Intel-based platforms was an added benefit.” —Tom Kuczynski, vice president, Information Technology, DC Water

The solution contains a dataset of 26,600 annotated images extracted from documented pipe inspection videos. A machine learning model built from these images was trained using TensorFlow* and optimized with the Intel® Distribution of OpenVINO™ toolkit for inference.

This training enables Pipe Sleuth to compare new pipe inspection videos with established anomalies. The solution can rapidly detect quality issues and eliminate the need for manual review and coding of underground sewer pipeline video scans (see Figure 1).

Figure 1. Sample anomalies and annotated images using the Pipe Sleuth solution.

Pipe Sleuth then generates a comprehensive inspection report, which can be integrated with commonly used asset management systems to automate maintenance work orders. Pipe Sleuth can also interface with geographic information software systems, which can map the precise locations where tasks must be performed. The software also makes it simple to retrieve and view specific scenes in pipe inspection videos.

In addition to optimizing the solution to work with the Intel Distribution of OpenVINO toolkit, Wipro also optimized it for Intel® Core™ i5 and Intel® Core™ i7 processors, and Intel® Xeon® Scalable processors. That enabled the solution to deliver gains of 32, 55, and 77 percent respectively.3

Inference time was also improved with a reduction of up to 80 percent using Intel Xeon processors with the OpenVINO toolkit, while not producing significant loss in model precision or accuracy.4 This optimization was a strong benefit for DC Water, which had already invested in Intel® processor-based servers.

The result: a solution that can perform image inferencing in real time, allowing pipeline inspections to improve in efficiency, consistency, and accuracy.

Furthermore, Pipe Sleuth enhances long-term monitoring by storing pipeline data and metadata. This makes it simple to look back at the history of specific problems and their locations, establish when and what maintenance actions or repairs were performed, and compare previous conditions with their current status.

Pipe Sleuth Benefits
Pipe Sleuth today supports detection of 50 anomalies targeted for wastewater utility infrastructure. Wipro plans to support additional anomalies as part of the product roadmap, which will enhance its business value.

Saves Time
Takes 10 minutes to analyze 60 minutes of inspection video and produce a report, which to do manually requires an hour and fifteen minutes.3

Reduces Scanning Costs
Saves up to 50 percent of the cost of anomaly detection, allowing utilities to shift spending from maintenance to capital improvement.5

Provides High Accuracy
Achieves a 90 percent accuracy rate while eliminating human errors due to fatigue and distraction and improves detection reliability by up to 20 percent.5

Increases Availability of Experts
Enables staff to focus on complex problems by using AI to perform routine inspection.

Optimizes Maintenance Decisions
Improves the process, allowing maintenance crews to work where they’re needed most.

Reduces Disruptions
Speeds analysis and prioritization so maintenance may occur before repairs must take place.

Improves Customer Service and Safety
Demonstrates commitment to customers by reducing the frequency of repairs and closures and damage to roads, public and private property, and the environment.

Delivers ROI
Utilities can experience up to a 350 percent ROI over a three-year period.5

Get Started Today
For more information download the white paper: Pipe Sleuth with Optimized Inference on Intel® Processors

Hear a podcast discussion on Pipe Sleuth with Deepak Dinkar, Senior Practice Manager, Wipro, and Emily Hutson, Senior Product Marketing Manager, AI Products Group, Intel Corporation

For other AI Wipro solutions, please contact Deepak Dinkar at deepak.dinkar@wipro.com

For Pipe Sleuth-related queries, please reach us at sales.pipesleuth@wipro.com and Thomas.Kuczynski@dcwater.com

Explore Related Products and Solutions

Intel® Xeon® Scalable Processors

Drive actionable insight, count on hardware-based security, and deploy dynamic service delivery with Intel® Xeon® Scalable processors.

Learn more

Intel® Deep Learning Boost (Intel® DL Boost)

Intel® Xeon® Scalable processors take embedded AI performance to the next level with Intel® Deep Learning Boost.

Learn more

OpenVINO™ Toolkit

Build end-to-end computer vision solutions quickly and consistently on Intel® architecture and our deep learning framework.

Learn more


インテル® テクノロジーの機能と利点はシステム構成によって異なり、対応するハードウェアやソフトウェア、またはサービスの有効化が必要となる場合があります。実際の性能はシステム構成によって異なります。絶対的なセキュリティーを提供できるコンピューター・システムはありません。詳細については、各システムメーカーまたは販売店にお問い合わせいただくか、http://www.intel.co.jp を参照してください。// 性能に関するテストに使用されるソフトウェアとワークロードは、性能がインテル® マイクロプロセッサーだけに最適化されていることがあります。SYSmark* や MobileMark* などの性能テストは、特定のコンピューター・システム、コンポーネント、ソフトウェア、操作、機能を使用して測定したものです。結果はこれらの要因によって異なります。製品の購入を検討される場合は、他の製品と組み合わせた場合の本製品の性能など、ほかの情報や性能テストも参考にして、パフォーマンスを総合的に評価することをお勧めします。詳細については、https://www.intel.co.jp/benchmarks (英語) を参照してください。// 性能の測定結果はシステム構成の詳細に記載された日付時点のテストに基づいています。また、現在公開中のすべてのセキュリティー・アップデートが適用されているとは限りません。詳細については、公開されている構成情報を参照してください。絶対的なセキュリティーを提供できる製品やコンポーネントはありません。// 記載されているコスト削減シナリオは、指定の状況と構成で、特定のインテル® プロセッサー搭載製品が将来のコストに及ぼす影響と実現されるコスト削減の例を示すためのものです。状況によって異なる可能性があります。インテルは、いかなるコストもコスト削減も保証いたしません。// インテルは、本資料で参照しているサードパーティーのベンチマーク・データまたはウェブサイトについて管理や監査を行っていません。本資料で参照しているウェブサイトにアクセスし、本資料で参照しているデータが正確かどうかを確認してください。// いくつかのテスト結果は、インテル社内での分析またはアーキテクチャーのシミュレーションあるいはモデリングで推定 / シュミレートされており、情報提供を目的として提供されています。システム・ハードウェア、ソフトウェア、構成などの違いにより、実際の性能は掲載された性能テストや評価とは異なる場合があります。