8 ways in which AI will change the development of product lines

Dealing with product lines is one aspect of systems engineering that could benefit from the rapid development of AI techniques. In this article, I explore this question together with Dr. Yang Li, Field Application Engineer and Consultant at pure-systems. Together we have identified 8 possible ways in which AI could bring us closer to this goal.

Update 2023: PTC buys PureSystems

It is exciting that Pure Variants has now been acquired by PTC. In contrast to other portfolio providers such as IBM, Siemens or Dassault, PTC now offers professional variant management.

Motivation

Today's customers demand products that are perfectly tailored to their needs. At the same time, the Product complexity and increasing. Both trends are leading to an increase in the number of product variants. The systematic co-development of several products in a product line is the focus of Product Line Engineering (PLE). However, many product companies have started their developments without PLE. However, they soon realize that they will need it sooner or later in order to master the increasing number of variants. Other companies are already successfully operating product lines and want to optimize their continuous development. They can all benefit from AI.

Below we present 8 ways in which AI can help companies overcome these challenges. We have grouped these according to the point in the product line lifecycle at which they are applicable.

Getting started with PLE with AI

Building a product line with feature-based PLE requires upfront investments, e.g. defining the scope of the product line, creating initial feature models with variability information, defining variants, etc. In the following, we describe some potential AI-supported use cases that could reduce the effort and costs of PLE.

In this article, we talk about the possible applications of AI techniques in the context of feature-based PLE. If you want to learn more about feature-based PLE, the ISO/IEC 26580:2021 Required reading.

Let's be honest: how many projects start with variant management in mind? Very few. Therefore, many companies already have a pile of documentation for existing products. These represent valuable input for the launch of a product line in the domain of their existing products. However, these legacy documents are often unstructured and difficult to reuse.

Few projects involve PLE from the outset.

There are many possibilities, Natural Language Processing (NLP) to process existing artifacts. For example, we can automatically analyze requirements specifications and product descriptions and extract features. Extracting keywords/phrases from the requirement specifications could be the first step to find the list of eligible features. Various technologies enable the automatic extraction of keywords/phrases, including:

• TF-IDF (frequency-based method), TextRank (graph-based method)
• Neuronal word embedding (e.g. BERT)
• Semantic roll labeling or Named Entity Recognition.

The high importance keywords/phrases identified using the above techniques (or a combination of these techniques) may represent certain functionalities that can be expressed as features. With the development of neural networks and deep learning, some of the above techniques have achieved higher accuracy in the general NLP datasets. This favors their application in the field of PLE. However, further analysis is essential to refine the feature list and establish reliable traceability between the extracted features and the existing artifacts. The Traceability enriches the domain knowledge of the features with additional documents and not with keywords/phrases with limited information, which is very helpful for both manual and automatic analysis.

It's one thing to have a feature, it's another to have the description. Once a feature has been identified, AI can help to describe it. To this end, it could compare similar feature descriptions in different documents and help combine them into a unified description. Or automatic text summarization techniques could be used to improve the management of all content.

Variability information is essential for PLE. At the same time, it is difficult to identify such information when product variants have evolved over years without systematic variant management. AI can alleviate the pain of such reverse engineering. It can help identify the exact/potential variability information, e.g. whether a feature is optional or mandatory, whether a feature works in combination with other features or clashes with other features, etc. AI can also support the structuring of such information in an easily understandable way for users.

To extract the relationships between features, the legacy artifacts associated with some features need to be further analyzed. AI can help group related features based on the functionality, semantic information or potential associations extracted from the corresponding legacy artifacts.

Useful techniques for the extraction of variability information are, for example, language modeling, similarity calculation, clustering or association rule mining.

Achieving a good structure of the features (i.e. the feature model) is a real milestone in the feature modeling phase, which depends on how similar the features are from the different perspectives, how strong the relationships are and what goals are to be achieved with the feature model.

Feature configurations for product variants are indispensable. Extracting feature configurations from legacy artifacts can significantly reduce the cost and effort of migrating to the product line. The legacy artifacts of similar products can also be considered variant-specific assets, but these legacy artifacts were not developed using a feature model-based method. AI can help re-establish the missing links between artifacts and features, which in turn can lead to the formulation of feature configurations for product variants.

Existing feature models are a prerequisite for success, because if you do not know information about your features in advance, it is impossible to find out the feature configurations from the legacy artifacts. Feature models capture the variability information of a product family in a structured way. To automate the extraction of configurations, you must first help the AI to understand the meaning of the existing features. That is, you need to digitize the domain knowledge in a form that allows us to easily process, analyze and understand the knowledge. To achieve this goal, we could combine the feature model with knowledge graphs/semantic networks that contain the domain knowledge. Furthermore, the initially extracted feature configurations can be analyzed using similarity computation and clustering algorithms to optimize the number of variants and the corresponding feature configurations.

AI reads the minds of your customers and stakeholders

It is crucial for every company that its products offer added value for its customers. In today's fast-moving world, however, it is more difficult than ever to get this right. The success of one product is no guarantee that the successor product will also be successful.

AI can help to incorporate customer input as early as possible and align it with the overall strategy of the product line.

Especially in the B2B environment, companies may develop a separate product for each customer based on their requirements. If you follow the PLE best practices, you already have a solid functional model for your product line. New functions are implemented in the context of the existing product line architecture. Artificial intelligence can help you to map the user requirements to your functional model for optimal customization. This can reduce the effort required to configure variants.

PLE best practices are the basis for the rapid implementation of new functions. In practice, however, this requires a lot of experience.

Consider, for example, a filling station where each customer has differently shaped containers, requirements for handling the contents (liquid? pellets?) and so on. The development of such a system typically involves multiple engineering disciplines, and many stakeholders are involved in developing the higher-level engineering assets for the product lines. These assets include superset requirements, superset test cases, superset source code, superset architecture, etc.

The ability to use AI to automatically or interactively map the existing functions to the corresponding existing engineering assets in multiple disciplines can therefore reduce the significant cost and effort required to develop a customized product.

The techniques discussed here are similar to those listed in points 1-3. However, these similar techniques can also be used for a different use case, namely analyzing the relevance or similarity between user needs/assets and existing features.

But where do these user needs come from? AI is already being used to process customer input for use cases that have nothing to do with PLE. These include analyzing support tickets or customer conversations, searching user forums for information, processing product telemetry data or observing sales behavior.

Today, ready-made AI systems extract valuable information from these data sources to find out, for example, which features receive more positive customer feedback and which features generate more customer complaints supported by automated sentiment analysis or facial emotion analysis. With regard to data protection, we must handle sensitive data such as videos or images that contain customers' personal data with care. Using this information to improve your product lines is just a small step.

An interesting variation of the above example is the identification of desired but non-existent functions. Sometimes customers articulate this themselves, for example in the form of function suggestions. Sales can also gain insights by analyzing lost sales. AI-supported data analysis can help you to automatically cleanse your data, extract interesting information and make predictions using pre-trained models. In this way, the missing features can be obtained using a data-driven methodology, for example by clustering similar ideas using topic modeling techniques. Using the approaches mentioned above, the AI transfers these new features to the existing feature model.

Understanding and optimizing the PLE value chain

The development of product lines is another point we need to consider. With the growth of product lines and the increase in the number of product variants, whether product lines can develop in a reasonable, controllable and predictable manner depends on the optimization of the product line structure, variability information and variants.

AI can play a role in the development of product lines to learn product line-related knowledge in the process of PLE and help make rational decisions.

As product lines evolve, a lot of data accumulates over time, e.g. variants, configurations, features and assets. AI techniques can be helpful in analyzing the large amounts of data associated with product lines and provide decision-makers with different perspectives.

It would be ideal if there was no product variability for each customer, but this is rarely the case in reality

To meet the needs of different stakeholders, variability may increase, leading to an exponential increase in the number of possible variants. The explosive increase in variability leads to an increasing workload on a large scale for various PLE activities. But perhaps not all the variability is necessary for the product lines, and we can optimize it. The Big Datathat you have at your disposal can be used to analyze your product lines with the help of AI.

For example, the structured domain knowledge in digital form (e.g. knowledge graphs) extracted and formulated from the big data can be considered as a central knowledge base that is helpful in automating the analysis process. In addition, we can use predictive analytics supported by machine learning and deep learning techniques. This includes approaches such as decision tree, linear/logistic regression and neural networks. This allows us to create predictive models that we train using historical data and known outcomes. The predictive models are then able to predict the results with the new data, which speeds up the analysis and optimization of product variants in an efficient and reliable way.

In particular, the combination of field usage data with variability information such as feature models and configurations can provide us with new insights with the help of ML. For example, we can recognize patterns between feature use and certain problems with products in productive use. These correlations are not easy to recognize in the huge amount of information generated by traditional data science techniques.

As mentioned earlier, AI can help you find variability in reverse engineering at the start of a product line. There is also the possibility of using AI to analyze big data related to the product lines, which can provide feedback to product portfolio managers or other stakeholders to further improve the product lines. For different PLE activities, several separate AI systems can be used for different activities - which is fine. However, it would be better if your AI performed these activities across the board to learn the interactions.

A comprehensive (end-to-end) use of AI for PLE topics is ideal

Why is end-to-end PLE intelligence so important? Because although each activity may be an independent task, the quality of the outcome of one activity influences the next activities, which in turn impacts the overall PLE. The generic pre-trained models may not be sufficient to formulate the end-to-end PLE intelligence, as the real-time intelligence is crucial for the end-to-end PLE intelligence. For example, the real-time changes of PLE activities and their effects should be learned to enable the development of the end-to-end PLE intelligence. Therefore, the design and implementation of end-to-end PLE intelligence is a very complex task that requires a combination of big data, complex deep learning algorithms and good software engineering.

What comes next?

Imagine a world where you can feed your system with unstructured data and get an optimized feature model with descriptions, parameterization and everything else. That would be the "Holy Grail" of AI in product line development.
Even if none of this may seem revolutionary at the moment, the use of these technologies holds enormous potential in the long term. It can make the difference between a company scaling and delighting its customers - or having to file for bankruptcy.

About Dr. Yang Li

Dr. Yang Li is a Field Application Engineer and also a consultant at pure-systems. He passes on his knowledge of product line engineering and pure::variants in tutorials, training courses and workshops to support customers on their way to systematic variant management.
Yang holds a PhD in computer science from the University of Magdeburg. He has focused on researching the application of artificial intelligence techniques to improve work efficiency in the field of product line development.

Image: Pixabay