Deep learning (DL) models are enabling a paradigm shift in how intelligence can be built into applications and real-world, daily life scenarios. Until recently, experimentation with deep learning models was largely limited to the scientific and research community. Now fueled by the abundance of data, computational power allowing for accelerated parallel processing, and the availability of software frameworks and models, start-ups to large enterprises are all adopting an AI-first strategy to derive actionable insights to improve and grow their business.
In this Part I of a multi-part blog post series, we will examine the top AI use cases in the manufacturing, telecom, and healthcare verticals.
AI Use Cases in Manufacturing
The intersection of internet of things (IoT) and AI results in the concept of smart manufacturing. This is an emerging field where cognitive techniques are leveraged to drive optimizations in areas such as asset management, supply chain management, fleet management, and order tracking.
The top AI use cases in manufacturing revolve around –
- Predictive maintenance: Ability to anticipate machine and equipment breakdown improves asset utilization and productivity. This includes both recognizing and forecasting failures ahead of time. Predictive maintenance results in reducing equipment downtimes and savings related to inspection costs.
- Example: ‘Predix’ is an industrial IOT platform (using AI) from GE for manufacturing that uses sensors to automatically capture all process steps and monitor complex machines. This platform incorporates complex event processing and machine learning (ML) at the edge.
- Yield enhancement: Reducing defects that cause products to be discarded has a direct impact on an organization’s bottom line. In the semiconductor industry, yield losses in production constitutes a big percentage of the total production cost. Using AI engines, companies are identifying causes of yield losses and locating yield detractors that can be avoided by changes to production processes.
- Example: Design for Manufacturability (DFM) is the process of designing products so they’re easy to manufacture. Scientists at the Iowa State University are developing a GPU-accelerated AI DFM decision-support framework to help designers optimize their CAD models to ensure manufacturability.
- Testing and quality optimization: Faster the products are screened for quality, higher is the throughput of a company. AI-enabled visual inspections are proving to be more accurate than human checks and operate at lower cost points. These algorithms rely on supervised learning to detect previously known defects and semi-supervised learning techniques to identify previously unknown defect types.
- Example: Bosch was able to achieve 35% reduction in test and calibration times in the production of hydraulic pumps for which they used ML models that would predict test outcomes and learn over time.
- Improvements in inventory management, better demand forecasting, and supply chain management are other big areas where AI techniques are being leveraged.
- Example: Honeywell uses an AI based demand forecasting model that takes differences and ratios of crude oil price index as inputs, they use this model for procurement, strategic sourcing, and cost management.
AI Use Cases in Telecom
At a high level, the two main drivers for the use of AI in telecom are reducing OpEx via automation and delivering a better customer experience. Per Tractica, telecom investment in AI is expected to reach $36.7B annually by 2025. The leading use case for AI telecom deployments are expected to be network operations monitoring and management, representing the lion’s share of AI spend over the period. Other key AI use cases center around virtual assistants for customer care, intelligent CRM systems, and cybersecurity.
The top AI use cases in telecom –
- Network optimization: Leverage AI to predict optimal connectivity for telecom networks, drive sophisticated network analysis and simulations, and intelligent network planning and drive optimizations.
- Chat bots: AI-powered automated chat applications to handle customer service inquires, to route customers to the agents in the appropriate departments, and to route prospective customers to the sales teams are key areas of focus in this function.
- Example: Spectrum uses an AI-powered virtual assistant called ‘Ask Spectrum’ to help customers with troubleshooting, account info, general questions.
- Example: CenturyLink has deployed a virtual assistant ‘Angie’, an AI-enabled software for sales and marketing that sends 30,000 emails each month and interprets the responses to determine hot leads.
- Speech and voice services: Service providers are looking to re-invent remote control units by powering them with speech recognition capabilities. This allows for a more personalized approach to selling content and services.
- Example: ‘XI Talking Guide’ voice-activated AI tool from Comcast that speaks, shows titles, channels, and time slots.
- Predictive maintenance: Fix hardware problems (cell towers, power lines, etc.) before they break, detect signals and break points that usually lead to failures. Use of drones to inspect phone towers is being adopted.
- Example: AT&T is developing a DL-based algorithm to fully automate drones to inspect its cell phone towers using AI for analysis of video data.
AI Use Cases in Healthcare
Healthcare is a process-oriented industry that offers an enormous opportunity to use AI to drive improvements, help meet unmet demand, and automate repetitive tasks. This is seen across R&D, patient care, medical imaging, and management tasks. Per Accenture’s recent research, AI applications in healthcare could save up to $150B annually by 2026.
The top AI use cases in healthcare –
- Robot-assisted surgery: These techniques can help analyze data from pre-operation medical records, physically guide instruments in real-time during a procedure, use data from actual surgical experiences to inform new surgical techniques. Expected outcome is to reduce errors and reduce patients’ length of post-surgery stay.
- Example: Mazor Robotics uses AI to aid minimally invasive surgical operations.
- Virtual nurse assistance: Voice and text apps trained to ask and handle preliminary health related questions. This enables wellness checks via voice and AI, reduces unnecessary hospital visits, assesses symptoms, and directs patients to the most effective care setting. The goal of these apps is to reduce the time nurses spend on patient maintenance.
- Example: Sensely offers an AI-powered nurse avatar ‘Molly’ that listens and responds to users.
- Diagnostics: Applying cognition to unlock vast amounts of health records, AI offers tremendous value in recognizing patterns across millions of scans in a short time with high accuracies. Areas of radiology and cancer research are prime targets.
- Example: Google AI’s LYNA (Lymph Node Assistant) reported 99% accuracy in detecting metastatic breast cancer.
- Administration: Non-patient activities consume an enormous amount of time for medical personnel. Use of voice-to-text apps to cut documentation time, improve quality reporting, and analyze thousands of medical papers using NLP to inform treatment plans saves time, increases patient care, and reduces inefficiencies.
Healthcare Case Study with ONTAP AI
As part of our research into the AI use cases across verticals, we ran a few use cases in each vertical on our ONTAP AI platform.
This is a health care related use case to classify images of breast cancer tumors. We used a data set of cell images from the University of Wisconsin, a CNN with 3 convolutional layers, and 2 fully-connected layers, all on one DGX-1, AFF A800, TensorFlow with data being served through FlexGroup volumes.
We were able to train the model to achieve 79% accuracy in identifying that a cell is benign and 92% accuracy that a cell is malignant within the testing dataset. These accuracies are of course limited to the small data set used but it showcases the art of the possible with AI and ONTAP AI.
AI applications across most verticals require some level of data orchestration between edge, core, and cloud, as a result seamless data management becomes important. Depending on the data source, size of data set, and cost points, organizations can choose to develop AI apps on public clouds or on-premises.
In Part 2 of this blog series, we will take a look at AI use cases in the Financials, Retail, and Government verticals.
For more information on artificial intelligence, please visit www.netapp.com/ai.