When industrial leaders consider AI-driven predictive maintenance solutions, most choose one of two paths:

1. Generalized platforms that enable do-it-yourself (DIY) Machine Learning model-building — like AWS SageMaker or Azure Machine Learning Studio.

2. Dedicated, industry-specific applications with pre-built data science models that offer AI-enabled insights.

The decision between the two is more complex than the typical app-versus-platform selection process. That’s in large part because, as it’s emerged, Industrial AI has developed along one of these two paths. Many industrial software companies operate from a place of either deep immersion in one specific industry or a generic approach to asset-intensive industries, and their products reflect this either-or legacy.

For example: if you take a look at the recent report from Gartner on asset performance management (APM) software, you’ll see this categorization of vendors as an “asset analysis product” or “APM platform.” That’s because, to date, most vendors have presented these two configurations as exclusive options to asset-intensive operators, service providers, and OEMs.

It’s a false and oversimplified choice — industrial leaders must be equipped with hybrid solutions that combine the benefits of both approaches and address the gaps in each.

Two Camps in Industrial AI

When a leader asks her team for recommendations on Industrial AI solutions, preferences often land in one of two camps: Chief Digital Officers and data scientists tend to like platforms, because they provide technical toys to play with, in-the-weeds-level control over eventual solutions, and likely more credit for business outcomes that the AI initiatives achieve. As generalized industrial platforms, they’re built with a birds-eye view toward enterprise-wide initiatives.

Asset managers, reliability engineers, and other operations leaders, on the other hand, usually want dedicated solutions. Like pre-Moneyball baseball scouts, they doubt algorithms can tell them something they don’t know about their assets — unless the software company selling them has industry credentials and its own in-house subject matter experts (SMEs). As industry-specific applications, they’re built with a narrower concentration on asset-specific maintenance activity.

In reality, both approaches carry significant downsides.

The greatest strength of generalizable AI/ML platforms, their configurability, is also their greatest weakness. In order to provide value for data scientists in sectors from marketing to manufacturing, they rarely offer the level of depth that accelerates speed to value in specific industries. For instance, a general platform cannot provide relevant additional information (such as benchmarks of peer customers’ asset or faults data), or a broad set of pre-built AI/ML models (like anomaly detection for a stamping press). And they might hamper additional, industry-specific plug-ins down the road because of short-sighted platform development.

Most Industrial AI applications, on the other hand, are limited by their depth and specificity. Heavy equipment companies own and operate broad sets of assets, and buyers and users alike prefer a single screen to manage and monitor everything in one place. On the part of operators, having multiple vendors becomes unwieldy and undermines enterprise-wide clarity and agility around asset classes. They’re pressed to pick and choose use cases for AI with the most determinable ROI rather than the broad deployment that can modernize maintenance and drive digital transformation.

Even sectors like power and utilities boast a diverse set of assets and use cases (e.g. renewable generation, transmission, thermal generation). This industry-specificity can require expensive customization to expand from a company’s core product, or a mix of point solutions that need to be checked individually to get a view of the whole enterprise.

Towards a Hybrid ModeI

As a result of these challenges to each approach, we’ve seen the Industrial AI sector — both in customer demand and vendor offerings — moving towards a hybrid model.

On a more technical level, this mixed approach combines data mastering and metadata-driven app development. Data ingestion isn’t just following a niche procedure because it’s from one system integrator or another. Instead, it’s a reusable and therefore scalable framework that accounts for the type of connection (API, FTP, etc.); component (e.g., a diesel engine, regardless of the asset it’s powering); data density and velocity; and available signals as mapped to a standardized, OEM-agnostic set of channels.

This process, when coupled with a framework and set of ML templates — like anomaly detection, failure prediction, and fluid analysis — speeds the configuration of models, often with the help of SMEs at the customer. It also makes application deployment for a variety of assets and similar use cases much easier, because implementation teams can edit a text JSON file instead of writing brand-new code. In addition, this hybrid approach enables the mixed approach to quickly incorporate contextual data sources like weather, as relevant to customer needs.

Uptake created a purpose-built platform with this strategy in mind, accelerating our ability to configure and deploy AI-driven predictive maintenance applications across a variety of sectors. Using our Digital Industrial Library, Uptake deploys pre-trained models to new assets, cleans and organizes enterprise asset data, and then fine-tunes our data science models to deliver insights that quickly realize value for our customers. We’ve honed this process of deploying models to new asset classes, and our work with General Motors iterated this framework to prevent failures on automotive stamping presses.

Check out this video for more on our work with GM:

While this shift in the market is in its early phases, it’s a strong indication that vendors are giving customers what they desire. The success of Industrial AI, and its ability to impact the bottom line of asset-intensive industries, rests on our ability to strike the optimal balance among configurability, specificity, usability, and knowledge.

Want more industrial insights?