Successfully using AI in the context of smart products

AI is becoming more and more of a commodity - we are familiar with this from our private lives (e.g. through ChatGPT or chatbots in customer service), but increasingly also in the B2B sector and in the field of smart products. In order to implement new use cases and project ideas successfully and efficiently, a structured, targeted approach is recommended, for example using the Data Science Life Cycle.

AI makes it possible

Smart products are inherently equipped with a certain basic intelligence without which they cannot fulfill their function.

With the addition of artificial intelligence, however, completely new use cases can be implemented that offer added value for manufacturers and operators - e.g:

AI-supported optimization of a machine's operating parameters - for quality assurance, for example, or generally for efficient, sustainable machine operation. AI opens up new possibilities here, as optimal operation requires both a deep understanding of the machine (which only the manufacturer has) and knowledge of the individual operating situation (upstream or downstream process steps, spatial environment of the machine, etc.), which only the operator has. It is generally not possible to bring these two areas of expertise together in a concentrated manner, but an AI-supported analysis can be the solution to this problem.
Predictive maintenance for products in customer use: This also involves evaluating and interpreting individual health, performance and process data from machines/products operated by the customer in real time - a complexity that is difficult or impossible to map without AI.

Both scenarios can be initiated by both the operator and the manufacturer and can be technically mapped both embedded and in the cloud.

AI is therefore an effective means of addressing complex problems - but a complex problem does not mean that the implementation must also be complex. So the question arises as to why, in practice, AI projects often fail in companies or only deliver inadequate results?

There is actually no reason for this. Because with a structured approach such as the Data Science Life Cycle and solid groundwork, AI projects can also be implemented in a targeted and efficient manner.

This is shown in practice

"A large part of an AI project is concerned with the groundwork. On average, around 38% of the time is spent on data preparation and cleansing. However, there are also key topics such as building an understanding of the business and data acquisition, which have to be carried out upstream in order to be able to start with the starting point - the data and its meaning - in the first place."

Data Science Life Cycle

Business understanding

In this phase, we clarify the basics: What is the use case or business case? Which problem is to be solved, which process is to be optimized? What data is required and is it already available or do technical requirements need to be created? Where should the data be processed - embedded on the device or in the cloud? At the end of this phase, we know what data is required for the use case, in what form and how it can be collected (with additional sensors if necessary).

Data Mining

This phase describes the data acquisition, i.e. the operational data collection. This involves not only the data from a smart product itself, but also - depending on the use case - the addition of process data or the creation of metadata - for example, when it comes to optimizing the operating parameters of a machine in the context of a manufacturing process.

Data Cleaning

This involves cleansing the data, removing errors, filling in gaps or deleting unnecessary data as a result of the data mining process. In order to successfully implement this step, a deep understanding of the data must be available. The main question is: How should incorrect or missing data be handled so that it does not affect the quality of the ML model later on?

Data Exploration

Phase 4 includes classic data analytics - this is about gaining a deeper understanding of the data. This includes correlations, recognizing anomalies and testing hypotheses (agreeing with previously formulated assumptions regarding the data situation), etc.

Feature Engineering

AI is used directly for the first time in feature engineering: variables are extracted in order to train models, new values such as variance, maxima or minima are created, which are later used in the models. Here in particular, it is necessary to draw on the know-how from business understanding so that relevant features can subsequently be created.

Predictive modeling

This phase involves setting up fully trained, AI-supported models that map the initially defined business case. There is often a misconception that this is where the greatest effort is required - but the fact is that many applications and libraries provide strong support and in some cases also automatically train different models and compare their performance (see AutoML).

Data Visualization

In the final phase, user interfaces such as dashboards or reports are built, which make the goals defined in the business case available to the end user in a user-friendly way. The main aim here is to finally check how the generated AI model can be used productively and what performance can be expected.

In practice, it will often be necessary to carry out a second data mining (with an improved data basis) after data cleaning or even to take another look at the business understanding or the desired business case - especially if it turns out that the goal cannot be achieved with the existing data and no additional data can be acquired through further sensor technology.

Again, the more careful the work in phase 1, the less need there will be for subsequent adjustments. BLUE-ZONE GmbH has extensive expertise and project experience, particularly in phases 1-3, from which our customers benefit in the long term.

According to Anaconda's Data Science Report, an average of 38% of the total costs of an AI project are incurred for data preparation and data cleansing alone. However, there are also upstream activities such as acquiring the data and gaining an understanding of the case. At the same time, this is where the course is set for how targeted and efficient the project will be in the future: If the foundation has been laid properly, less reworking, corrections etc. will be required in the later phases.

As described above, the success of an AI project and the overall effort involved depend to a large extent on a clear foundation being laid in phases 1 and 1-3.

Our offer for you

Invest 60 minutes in a non-binding innovation exchange to gain an initial basic understanding of how to set up an AI project and lay the foundations for the desired business case - and benefit from efficient and targeted project progress in the future.

Your contact person

Andreas Lehner, MSc

Head of Innovation, Sales

blue-zone GmbH

T +43 7236 78500-25

andreas.lehner@blue-zone.at

Contact form

Our workshops

View now