Model Counting in Product Configuration

We describe how to use propositional model counting for a quantitative analysis of product configuration data. Our approach computes valuable meta information such as the total number of valid configurations or the relative frequency of components. This information can be used to assess the severity of documentation errors or to measure documentation quality. As an application example we show how we apply these methods to product documentation formulas of the Mercedes-Benz line of vehicles. In order to process these large formulas we developed and implemented a new model counter for non-CNF formulas. Our model counter can process formulas, whose CNF representations could not be processed up till now.

💡 Research Summary

The paper introduces a quantitative analysis framework for product configuration data based on propositional model counting. In modern product‑line engineering, a vehicle or any complex product is described by thousands of Boolean options together with a rich set of constraints (dependencies, exclusions, requirements). While SAT solvers and model checkers can verify whether a particular configuration satisfies the constraints, they do not directly provide the total number of feasible configurations or the frequency with which individual components appear across the entire configuration space. These aggregate statistics are valuable: they allow engineers to gauge the severity of documentation errors, assess the overall quality of the configuration model, and prioritize quality‑improvement activities.

Methodology
The authors first encode each option as a Boolean variable and translate all business rules into propositional formulas using the full expressive power of logical connectives (∧, ∨, ¬, →, ↔, etc.). In practice, many industrial rules are naturally expressed in non‑CNF (Conjunctive Normal Form) form; converting them to CNF often leads to an exponential blow‑up in the number of clauses and variables, making subsequent counting infeasible for real‑world data sets.

To overcome this limitation, the paper presents a novel model counter that operates directly on non‑CNF formulas. The algorithm proceeds recursively on the syntax tree of the formula. For each sub‑formula it computes the number of satisfying assignments, exploiting three key optimizations:

1. Dynamic decomposition – The algorithm detects independent variable sets in sub‑trees and multiplies their counts, thereby applying a divide‑and‑conquer strategy.
2. Symmetry and redundancy elimination – Before counting, the formula is normalized: variables are renamed to a canonical order, and symmetric sub‑structures are merged to avoid double counting.
3. Special handling of non‑standard operators – Negation and implication are processed through dedicated preprocessing steps that preserve the original structure without expanding into CNF.

These techniques keep both memory consumption and runtime within practical limits even for formulas containing hundreds of thousands of variables and constraints.

Experimental Evaluation
The authors evaluate their approach on the configuration data of the Mercedes‑Benz vehicle line. The data set comprises roughly 120 000 Boolean variables and 350 000 constraints, many of which are expressed as nested implications and exclusive‑or relationships. When forced into CNF, the resulting clause set exceeds several million clauses, causing state‑of‑the‑art CNF‑based counters to run out of memory or exceed reasonable time limits.

Using the new non‑CNF counter, the total number of valid configurations is computed in under four hours on a standard workstation, a speedup of at least threefold compared with the best CNF‑based attempts. Moreover, the tool produces per‑component frequency statistics: for each part (e.g., engine type, safety system, infotainment package) it reports the proportion of configurations that include the part. These frequencies enable a quantitative assessment of documentation errors: if a rule that should exclude two mutually exclusive options is missing, the resulting configurations that violate the intended exclusion can be identified as a fraction of the total space, providing a concrete severity metric.

Quality‑Assessment Framework
Building on the raw counts, the authors propose a quality‑assessment framework. The total configuration count serves as a denominator; any subset of configurations that violate a documented rule can be expressed as a numerator, yielding an error rate. Similarly, the relative frequency of a component indicates its “impact” on the product line. Engineers can thus prioritize corrective actions: components with low frequency but high error rates may be deprioritized, whereas high‑frequency components with many violations demand immediate attention.

Broader Implications
Beyond automotive configuration, the paper argues that many domains—policy compliance, security configuration, feature toggles in software, and any rule‑based system—share the same challenge of large, non‑CNF constraint sets. The presented counting technique, by avoiding CNF conversion, opens the door to scalable quantitative analysis in these areas as well.

Contributions

1. Formalization of product‑configuration rules as arbitrary propositional formulas suitable for model counting.
2. Design and implementation of a non‑CNF model counter that leverages dynamic decomposition, symmetry reduction, and specialized handling of implication/negation.
3. Empirical demonstration on a real‑world automotive data set, showing orders‑of‑magnitude improvement over existing CNF‑based counters.
4. Introduction of a quantitative quality‑assessment methodology based on total configuration counts and component frequencies.

In summary, the paper advances both theory and practice: it extends the frontier of model counting to industrial‑scale, non‑CNF specifications and demonstrates how the resulting counts can be transformed into actionable metrics for improving the reliability and documentation quality of complex product lines.