Using Knowledge Graph Embedding for Fault Detection

Automotive manufacturers are under stressful timelines as they shift their focus from internal combustion engines (ICE) to electric (EV) and hybrid-electric vehicles (HEV). The demand for this rapid change is crucial meet a growing consumer market. New manufacturing challenges coupled with can lead substantial safety risks consumers well financial liability automakers, especially when recalls happen. resulting misplacement, misalignment, or defective assembly of any the components connectors result in critical even fatal outcomes consumers. Recent findings reported by CNBC revealed that had cost automakers billions dollars (Kolodny 2022). recalling an EV far outweighs ICE. For instance, Ford Kuga plug-in HEV re-calls costs about $19,000 per vehicle, contrast typical ICE vehicle recall averages around $500 (Isidore Vales-Dapena Furthermore, rate has been higher. China’s was approximately 6.9% its total sales volume (Hao et al. 2021).Automakers highly motivated prevent automotive implementing employing several preventative measures. IoT sensor-based fault detection systems, those camera capabilities, have used detect defects during production processes. Industry 4.0 standards (Garofalo 2022) adopted, particularly companies employ autonomous process.A issue vision systems limitations, where only analyze observe end without analyzing relationships possible underlying connections other components. these reveal whether given component missing connected correctly another component. Simply relying on machine examining isolation, uncontrolled environments, becomes difficult reliable, not mention extremely de-manding computational power needed processing.The motivation research work present alternative perspective employs collective view components, represented networked graph, knowledge graph (KG) we hypothesize ability be effective data search faults.KGs collection real-world fact triplets structured form (head, relation, tail) (Hogan Fundamentally, KGs expressed nodes represent sub-components, edges indicate relationship between two adjacent Hence, KG effectively map interconnected after manufacturing. Researchers demonstrated usefulness Knowledge Graph Embedding (KGE) potential solution detection, it advance driving solutions (Bosch Global 2022).This aims at building testing effectiveness detecting faults custom dataset. We implement Completion (KGC) algorithm compare different models. measure Mean Reciprocal Rank (MRR) Hits@K evaluate based various KGE approaches Our our experiments pave new pathway car makers, allowing feasible comprehensive system framework. By combining state-of-the-art models first-hand case study involving dataset (EV-KG), solidifies future KG-related field opens numerous opportunities further development application industry.Link prediction graphs accelerated recent years through works publications, KGE-based methods like RotatE (Bollacker 2008), which allows more accurate efficient (or edges) entities nodes) graph. Specifically, integration link enables analyzed simply individual but made up such vehicle.In method, first embark EV-KG develop all physical relations drawn domain experts manufacturer documentation. A dictionary file built each defined Next, RDF format generated validation. This using (battery_positive_connection, terms-nal_of, battery_cell). take pre-process randomized split into three distinct sets: training dataset, validation Each individually model phases (training dataset) evaluation (validation datasets). score function give candidate triples. represents distance nodes, thus, similar ranking metric, lower score, better. were conducted high-performance cluster compiled specifically vehicle’s layout, factoring parts faulty, will perform experiment. contains 1378 2200 edges, 15 unique relations. tested RotatE, HRotatE, pRotatE, DistMult, ComplEx, TransE modes (Sun al, 2019), found achieved best overall 0.922 [email protected] studying feasibility KG-based system, heavily emphasizes need making environment. Just important accuracy since also affect losses if, example, there too many false negatives positives. One goals make sure choice comes provide early before hands consumer.