New DaRT Model Predicts How Long It Takes to Repair Everyday Appliances

Repair has quietly become one of the most consequential words in modern manufacturing. As the European Union advances its right-to-repair agenda and repairability scores begin to shape consumer choices across the continent, the question of how easily a product can be taken apart and put back together has moved from the workshop bench to the policy arena. Now, a team of researchers at TU Delft has unveiled a new tool that promises to make those judgments faster, fairer, and far more broadly applicable. The tool, called DaRT, for Disassembly and Reassembly Time, was described in the Journal of Industrial Ecology and is built on an unusually rich foundation: more than 10,000 timed repair actions performed by professional repairers on real household products.

The problem DaRT addresses is deceptively simple to state but notoriously difficult to solve. Repairability scoring systems, including the French repair index and the European Union’s own assessment frameworks, need a reliable way to measure how hard it is to open a device and reach the parts that most often fail. The dominant approaches have each carried significant drawbacks. The so-called step method, codified in the EN45554 standard, simply counts the number of disassembly steps, such as tool changes and component removals. It is easy to apply, but previous research has shown a low correlation between step counts and the actual time a repairer spends working, which makes it a poor proxy for real-world repair effort.

The alternative is a family of proxy-time methods, most prominently the ease of Disassembly Metric, or eDiM, and the iFixit proxy time system. These estimate how long disassembly takes by modeling the individual actions involved, such as unscrewing a fastener or prying open a snap fit. Studies have found that these time-based approaches represent the ease of disassembly considerably better than step counting, because time is what matters most both to professional repair services and to consumers attempting a fix at home. Yet eDiM and iFixit proxy times only cover disassembly, ignore reassembly entirely, and were built around a limited library of fastener types drawn mainly from information and communication technology products such as laptops, monitors, and televisions. eDiM is also demanding to use, requiring time-consuming calculations, frequent updates, and navigation of a bewildering combinatorial space of tools and connectors.

The Delft team, led by Sagar Dangal with colleagues including Linda Ritzen, Alma van Oudheusden, Jeremy Faludi, and Ruud Balkenende, set out to close that gap with a model grounded directly in observed repair practice. Working within the Horizon 2020 PROMPT project, they selected four product categories through a multi-criteria analysis: washing machines, vacuum cleaners, smart televisions, and smartphones. These were chosen for their ubiquity in homes, their varied mechanical and electronic complexity, and their high repair frequency. The researchers analyzed 12 washing machines and 12 vacuum cleaners through full disassembly and reassembly, 7 televisions for prying actions, and 35 smartphones for adhesive-removal actions, spanning different brands and price ranges to capture the diversity of real product design.

The empirical protocol was rigorous. Professional repairers from iFixit and the Austrian repair organization RUSZ, each with more than five years of experience, performed the disassemblies in a realistic repair environment with proper lighting, space, and tools within reach. Power tools were excluded to keep the protocol simple. Each product was taken apart to its individual components, stopping at permanent fixtures such as soldering, welding, or thermal molding, and then reassembled. The repairers described every action aloud, noting the tool used, the force applied, the fastener type, and the visibility of each fastener. The entire process was filmed from front and top views, and the products were then disassembled and reassembled a second time by the same person, with timings extracted from the second video to eliminate learning-curve distortions. In total, 10,569 datapoints were collected.

The statistical treatment favored robustness over sophistication. The team used median values rather than means, because medians resist the influence of outliers and skewed distributions, and they grouped actions whose medians fell within each other’s interquartile ranges. Interestingly, an initial analysis using the Mann-Whitney test occasionally flagged statistically significant differences that were not meaningful in practical terms, so the researchers opted for the more pragmatic interquartile criterion. The resulting findings are illuminating. For most actions, including positioning tools and the majority of screwing and unscrewing tasks, disassembly and reassembly times are similar enough to share a single proxy value, an assumption long embedded in scoring systems but never before demonstrated empirically across such a broad product range.

But the data also revealed where that assumption breaks down. Reassembly takes significantly longer than disassembly whenever precise positioning or alignment is required: grabbing and placing screws or components is slower than removing them, routing cables takes more time than pulling them out, and plugging in a hose takes longer than unplugging it. One striking exception runs the other way: tilting a heavy product like a washing machine takes longer than tilting it back, because tilting means working against gravity. The analysis also showed that some actions differ by product size. Grabbing and positioning tools, handling hoses, and turning the product all took significantly longer on washing machines than on vacuum cleaners, largely because washing machine hoses must be watertight and withstand higher pressure, and the machines themselves are far heavier. The model therefore distinguishes between larger and smaller appliances for those three actions.

The grouping analysis produced further simplifications. Positioning a tool turned out to take roughly the same time regardless of tool type, allowing all such actions to be merged into one group. Screw and wrench timings were largely consistent across driver and wrench sizes, with a few exceptions, and the clearest dividing line was the number of turns: screws requiring more than 15 turns deviated significantly from the rest and were treated separately. Force classifications also needed adjustment. While existing standards distinguish low, medium, and high force using the MOST system, the researchers found that repairers could not reliably separate fingertip-level from wrist-level forces in practice, though the difference between medium and high force was clear. Putting away tools, screws, and components likewise could not be grouped, since components demand multi-dimensional alignment while screws need only a single slot.

Validation was left to independent testing bodies, including the VDE Testing and Certification Institute and the Müller-BBM Group, who dismantled four washing machines, five vacuum cleaners, and five smart TVs and compared actual times for priority parts, such as washing machine pumps and door locks, vacuum cleaner motors and cord reels, and television main boards and power supplies, against DaRT predictions. The correlation was remarkably strong, with an R-squared of 0.98 overall, and correlations of 0.99 for shorter tasks and 0.96 for longer ones. Usability testing showed the assessment took roughly three times longer than simply disassembling a product, with most of that time spent creating the disassembly map, a step the simpler counting method also requires. Compared with eDiM, DaRT values agreed within about 30 percent for most actions, though eDiM tended to underestimate tasks such as prying snap fits, unscrewing short screws, and unplugging wires, plausibly because it was derived from standardized assembly-line motion timings rather than the messier realities of repair.

The researchers are candid about the trade-offs. Broad distributions for actions like prying and loosening reflect genuine architectural differences between products, and for unusually simple or complex devices DaRT may under- or overestimate the true effort involved. Unique tasks, such as removing a washing machine door seal, are handled through a flexible complex-action input that lets users enter an approximate time without distorting rankings within a product category. Future work will extend the model to furniture and cars, add desoldering timings, and incorporate the power tools that are increasingly common in repair shops. For now, DaRT strikes what its creators describe as a favorable balance between accuracy and ease of use, outperforming step counting while being far more practical than eDiM. If repairability scores built on it steer manufacturers toward screws over glue and accessible parts over sealed assemblies, the humble stopwatch may prove to be one of the circular economy’s most powerful instruments.

Subject of Research: A data-driven proxy time model for assessing disassembly and reassembly times in product repairability

Article Title: Modelling disassembly and reassembly times (DaRT) for assessing repairability

Article References: Dangal, S., Ritzen, L., van Oudheusden, A., Faludi, J., & Balkenende, R. (2026). Modelling disassembly and reassembly times (DaRT) for assessing repairability. Journal of Industrial Ecology. https://doi.org/10.1007/s44498-026-00150-9

Image Credits: AI Generated

DOI: 10.1007/s44498-026-00150-9

Keywords: repairability, circular economy, disassembly time, reassembly, DaRT model, right to repair, household appliances, proxy time methods, eDiM, design for repair, electronics, sustainability