In the context of evolving digital innovation ecosystems, the assessment of technology progress remains a critical challenge for collaborative R\&D management. This paper evaluates the technological evolution of two distinct Digital Twin pilots, a hardware-integrated system and a software-centric operational platform, developed within a large-scale sectoral innovation pact in the textile industry. The research proposes a transition from traditional, linear maturity models toward a multidimensional Readiness Framework to better capture the functional value delivered in complex industrial environments.
The study follows a two-phase assessment design, comparing theoretical maturity projections with functional readiness profiles supported by verifiable technical evidence. The results demonstrate that while traditional maturity models provide essential strategic roadmaps, they often harbor a physics-based bias that overlooks the specialized value of software-centric solutions. Specifically, the findings reveal a functional jump in prescriptive intelligence that delivers immediate business value despite the absence of foundational visual representations.
By validating an evidence-based Readiness Framework, this research provides innovation managers and stakeholders with a robust diagnostic tool to bridge the gap between scientific maturity and market-oriented readiness. The study identifies functional nonlinearity as a hallmark of digital transformation trajectories, offering a standardized approach for technology assessment in collaborative R\&D consortia. Ultimately, this framework supports more informed decision making in digital ecosystems, ensuring that technological progress is measured by its operational impact and strategic readiness rather than just hierarchical complexity.

The preservation of classic vehicles demands an inventory discipline that ad hoc, memory-based warehouse practices are structurally incapable of providing. This paper presents a web-based inventory management system tailored to the classic car restoration domain, in which large vision models are embedded directly into the part cataloging pipeline to automate what has historically been a manual, error-prone process. Upon vehicle disassembly, the system automatically segments photographed components and spatially anchors each part to its original mounting location on a generated three-dimensional vehicle model, while barcode integration maintains a continuous physical-to-digital link throughout the warehouse lifecycle. The architecture is designed as an interoperable module within a broader digital restoration ecosystem, interfacing with an existing Charter of Turin process management platform and a facility-wide Digital Twin. The system was empirically validated through an on-site case study at a professional restoration facility, where it achieved complete component digitization and eliminated the recall errors observed in expert-based identification baselines. These results establish that vision model-driven digitization is a practically viable and cost-effective mechanism for introducing rigorous process improvement into small-scale heritage industrial environments.

As the complexity of cyber-physical systems grows, ensuring the internal consistency of large-scale requirements specifications becomes increasingly critical, yet prohibitively labor-intensive. Traditional manual pairwise cross-referencing struggles to scale with expanding requirements spaces, and existing natural language processing approaches typically reduce conflict detection to binary classification, failing to capture the graded spectrum of incompatibilities that characterize real-world engineering specifications. This paper presents a graded conflict detection approach based on a fine-tuned MiniLM cross-encoder architecture that formulates requirements conflict detection as a continuous regression problem, predicting pairwise conflict scores on a [0,1] scale. To address the absence of publicly available datasets with fine-grained conflict annotations, we developed a controlled synthetic generation pipeline that produces requirement pairs with known semantic relationships across compatible, partially conflicting, and strongly contradictory categories. Evaluation on a held-out test set yields a Pearson correlation of 0.883, a Spearman correlation of 0.863, and a mean absolute error of 0.12, demonstrating that the model effectively captures semantic relationships and preserves conflict ranking. A systematic qualitative analysis reveals that while the model excels at detecting explicit contradictions and clearly compatible requirements, it exhibits specific limitations in numerical constraint reasoning, modality sensitivity, and partial conflict detection. These findings highlight the boundary between semantic similarity modeling and logical reasoning in transformer-based architectures and inform concrete directions for future improvement through targeted data augmentation and hybrid neural-symbolic approaches.

Chatbots are increasingly used in customer service across different industries. Despite their widespread adoption, the academic literature still lacks a structured and comprehensive overview of the challenges organizations face when deploying them. This paper addresses that gap by developing a taxonomy of chatbot deployment challenges based on a systematic literature review following established standards. The challenges were coded and synthesized using qualitative content analysis, resulting in 23 distinct challenges which were grouped into seven aggregate dimensions. The findings show that these challenges are closely interconnected, with technical shortcomings increasing user frustration, privacy concerns undermining trust, and resource constraints limiting an organization’s ability to address underlying issues. The resulting taxonomy provides practitioners with a structured basis for identifying risks and managing expectations, while offering researchers a foundation for future empirical validation across industries and organizational contexts.