DisAssembly Scenario Learning Factory Dataset (DASCEN-LF)

This repository contains a dataset from a learning factory, which was collected using the flexible disassembly demonstrator developed by Jordan et al. (2025).

📊 Dataset Overview

• 6 scenarios | 60 production orders (toy cars) | 390 components tracked
• 3 to 5 workstations | Line & Workshop layouts | Manual & Automated processing
• 4 quality types (HD, TL, SA, RD) with variable disassembly depths (4-8 target components per quality)
• 18 unique component types tracked (9 target, 5 intermediate, 4 remaining)

Table of Contents

• Contact
• Introduction
• Dataset
• Dataset Details
• Cross-Scenario Data Analysis
• Citation
• License
• References

Contact

Contact Details

Corresponding author: Patrick Jordan
Institutional email: patrick.jordan@iwb.tum.de
Personal profile: patrick-jordan.github.io

Useful Links

• Visit our other repositories
  Explore more tools and resources from our research institute.

• Visit our institute for more information
  Learn more about our research and ongoing projects.

Introduction

This repository contains a dataset from six disassembly runs carried out in January 2025 at the Smart Production Lab (SPL) at the Institute for Machine Tools and Industrial Management (iwb) at the Technical University of Munich (https://iwb-spl.de/). The dataset provides the complete disassembly process data for 60 remotely controlled electric toy cars (RC cars).

A disassembly scenario combines two elements: a system design scenario (layout and workstation configuration) and a product scenario (incoming products and disassembly depths). Since these decisions are interdependent, multiple scenario combinations are possible. Comparing these scenarios is necessary to determine the economically and ecologically appropriate disassembly approach (Jordan et al. 2024) The six scenarios presented in this repository include system designs with three to five workstations, one to four product types per product portfolio, and various layouts. This provides diverse material flow data. The dataset for each scenario is divided into two types: gathered raw event data and pre-processed data, including an initial analysis of the performance of each disassembly scenario.

Research Context

The data was collected in the Smart Production Lab (SPL) at the Institute for Machine Tools and Industrial Management (iwb), Technical University of Munich. The SPL is a learning factory for lean, digital, automated, and circular manufacturing. It offers various disassembly system designs that represent different scenarios, building upon the infrastructure described in Jordan et al. (2025). As outlined in Jordan et al. (2025), the demonstrator facilitates the comparison of real-world data with synthetic simulation data. This enables the validation of disasembly simulation models while simultaneously exploring different disassembly system designs and the impact of the disassembly depth on the system performance. Beyond the proposed use for the validation of flexible disassembly simulation models, the data could be used for various purposes, such as comparing different layouts, benchmarking process mining techniques, and analyzing resource allocation policies in diverging material flows.

Related Research Work

The dataset is an extension of the initial testing phase of the demonstrator, as described in the following article:

"Demonstrator-based implementation of an infrastructure for event data acquisition in disassembly material flows"
Patrick Jordan, Lasse Streibel, Niklas Lindholm, Wesal Maroof, Susanne Vernim, Laura Goebel, Michael F. Zaeh
Proceedings of the 58th CIRP Conference on Manufacturing Systems 2025

For more details, see the published article (Link)

Experimental Design

A disassembly graph containing over 60 components was used to design four quality levels for the cars: Hail Damage (HD), Rear Damage (RD), Total Loss (TL), and Shock Absorber defects (SA). For each quality level, disassembly sequences were defined based on the graph structure. These sequences were subsequently translated into quality- and scenario-specific work instructions, which provided operators with a step-by-step disassembly plan and routing information for each production order.

Each disassembly scenario was executed for approximately 40 minutes with one operator per workstation. Each operator managed the material flow between workstations based on the information provided in the work instructions. An additional operator removed the final disassembled components from the shop floor. The logistics data after the last disassembly step cannot be used because components were randomly removed from the final buffer location (B-Transfer), making the timing data unreliable. From the solution elements of the demonstrator (Jordan et al. 2025), the assignment app was used to ensure that each incoming production order was linked to a single tracking device. During the disassembly process, the inheritance app was used to ensure that each component was linked correctly to the original production order. The collision zones were used for the automatic timestamp capture (event data), and the central database was used for recording that event data. Instead of utilizing the work instructions app, manual work instructions were provided to manage the material flow. For more information about the demonstrator, please refer to Jordan et al. (2025).

Data Quality Considerations

⚠️ Known Limitations:

• The signal-to-noise ratio varies due to the system's location-based tracking, which creates duplicates in timestamps. For example, a product may enter and exit a workstation multiple times if it is too close to the edge of the virtual collision zone. For more detailed information about the collision zones, please refer to Jordan et al. (2025).
• It has been observed that operators have, on occasion, deviated from the work instructions. For instance, operators disassembled multiple components simultaneously, not sequentially, as defined.
• Sometimes, timestamps had to be calculated from averages when events were not properly captured.
• Due to the aforementioned limitations, extensive manual data cleaning was necessary. This process may have introduced errors, such as selecting incorrect timestamps for certain activities.

Dataset

Dataset Overview

⚠️ Experimental Design Note: The experimental design prioritized operational diversity over performance optimization in the disassembly scenarios. The unbalanced workstation assignments and varied material flows are intentional. The objective of this work was to create diverse disassembly data, enabling the analysis of different system behaviors and to validate the simulation framework.

The detailed documentation, data, and analysis for each disassembly scenario can be found in the respective folders:

Scenario 1: Line Layout - Mixed Portfolio

• System configuration: 3 workstations, line layout, automated processing
• Product portfolio: RD (60%), TL (20%), SA (20%)
• Documentation: scenario_01_documentation.md

Scenario 2: Workshop Layout - HD Portfolio

• System configuration: 4 workstations, workshop layout, manual processing
• Product portfolio: HD (100%)
• Documentation: scenario_02_documentation.md

Scenario 3: Line Layout - HD Portfolio

• System configuration: 5 workstations, line layout, automated processing
• Product portfolio: HD (100%)
• Documentation: scenario_03_documentation.md

Scenario 4: Workshop Layout - HD Portfolio

• System configuration: 5 workstations, workshop layout, manual processing
• Product portfolio: HD (100%)
• Documentation: scenario_04_documentation.md

Scenario 5: Workshop Layout - Mixed Portfolio

• System configuration: 5 workstations, workshop layout, automated processing
• Product portfolio: HD (70%), TL (10%), SA (10%), RD (10%)
• Documentation: scenario_05_documentation.md

Scenario 6: Line Layout - Mixed Portfolio

• System configuration: 5 workstations, line layout, manual processing
• Product portfolio: HD (30%), SA (30%), RD (30%), TL (10%)
• Documentation: scenario_06_documentation.md

File Organization

Each scenario contains:

• Raw CSV data: Original tracking data with all noise and intermediate events
• Excel analysis file: Processed data with validation columns, extracted sequences, and analysis
• Material flow diagrams: Visual representation of the material flow and system layout (not to scale)
• Disassembly sequences: Visual representation of the disassembly process for each quality type
• Work instructions (WI): Quality-specific work instructions including production order information, delivery times, and step-by-step disassembly procedures. Note that each scenario has 10 production orders, but work instructions are grouped by quality type (e.g., scenario 1 has 10 POs but only 3 WI files for RD, TL, and SA quality types)

Raw Data - CSV Data Fields

Table 1 provides an overview of the structure of the raw CSV data fields utilized for the tracking events, along with the translation of the original German column names. The raw data files contain tracking events with timestamps, locations, and component identifiers.

Table 1. CSV data field structure

Column	Type	Description	Example
fauf	String	Production Order ID	PO-01
materialnummer	String	Component identifier	Car-01, BOSP
menge	Integer	Quantity	1
arbeitsstation	String	Workstation/location	WS-01, B-Transfer
eventtyp	String	Event type	Generation, Entry, Exit
zustand	String	Quality condition	Standard, HD
zeitstempel	Timestamp	ISO timestamp	2025-01-03 08:56:00.719+01
tag_id	String	Identifier of UWB Tag	84b31b00000b807d

Excel Analysis Structure

Table 2 lists the eight standardized sheets included in each Excel analysis file. Each sheet serves a specific purpose in the data processing and analysis pipeline, from raw event logs to final performance indicators.

Table 2. Excel sheet structure and purpose

Sheet	Description	Purpose
raw_data	Original tracking events	Unprocessed event log
delivery_times	Manual arrival records	Start time documentation
preprocess	Data validation and cleaning	Event filtering with OK/SKIP flags
extraction	Clean event sequences by PO	Final material flow per order
workstation_timeline	Workstation-based event ordering	Utilization analysis
process_calculation	Time indicators and KPIs	Performance calculations
analysis	Configuration and performance summary	Final indicators and KPIs
timestamp_calculation	Missing value estimation	Gap filling using averages

Component Dictionary

Table 3 lists all component codes and descriptions, which are standardized across all scenarios. For detailed information about the components and the material flows, please refer to the sequence files of the respective scenario.

Table 3. Component codes and descriptions

Code	Component	Quantity	Weight (g)[1]	Value (Euro)[2]	Destination	Notes
Car	-	1	1360	230	-	-
Target Components
BOSP	Body & spoiler	1+1	77	44	Recycle	-
BAT	Battery	1	396	42	Reuse	-
RT	Rear tires	2	142	26	Recycle/Reuse	-
FT	Front tires	2	97	24	Reuse	-
FAXS	Front axis with shock absorbers	1	155	50	None (intermediate assembly)	Consists of front axis, SSA[3]
SSA	Small shock absorbers	2	15	21	Recycle	Consists of: damper spring, damper housing, damper housing cap
FAX	Front axis	1	140	30	Remanufacture	-
CHS	Chassis	1	414	56	Reuse	Consists of chassis and power steering; only HD[4]
CORE	Chassis with systems and engine (excluding rear axis)	1	600	110	Reuse	only RD[4]
BSA	Big shock absorbers	2	15	21	Remanufacture	Consists of: damper spring, damper housing, damper housing cap
Intermediate assemblies[5]
CRAXS	Chassis with rear axis and shock absorbers	1	950	-	None (intermediate assembly)	only HD
CREW	Chassis, remaining systems, engine, and wheels	1	887	-	None (intermediate assembly)	only TL
RAXS	Rear axis with motor and big shock absorbers	1	540	-	None (intermediate assembly)	only HD, RD
CSEWF	Chassis with systems, engine, and front wheels	1	1140	-	None (intermediate assembly)	only RD
Remaining Parts[6]
RAX (HD)	Rear axis	1	500	50	Recycle	Consists of rear axis, engine, gear cover, gears; only HD
RAX (RD)	Rear axis	1	500	50	Repair	Consists of rear axis, engine, gear cover, gears; only RD
CRE (TL)	Chassis, remaining systems, engine	1	648	94	Recycle	only TL
CSEB-NABS (SA)	Chassis, systems, engine, body	1	1090	138	Recycle	only SA

Notes:

^[1] Component weights are based on manual weighing.

^[2] Component values are estimated based on the new value. For assemblies, the approximate sum of the components contained within is used.

^[3] FAXS is disassembled into FAX and SSA. In some cases (e.g., scenario 2), timestamps for this split were gathered manually by the operator.

^[4] CHS and CORE represent physically distinct assemblies corresponding to different quality types (HD and RD, respectively). However, the disassembly tasks performed on both components are identical. Process times are therefore reported as CHS/CORE.

^[5] These assemblies represent temporary intermediate states during disassembly. They are not tracked as separate entities in the dataset.

^[6] Remaining parts represent assemblies that retain the tracking tag from the original car and are not fully disassembled.

Disassembly Strategies

Table 4 summarizes the four quality-dependent disassembly strategies. Different disassembly strategies were implemented based on simulated quality conditions, resulting in varying disassembly depths and component recovery patterns.

Table 4. Quality-dependent disassembly strategies

Code	Quality type	Strategy	Target Components	Count	Remaining Assembly	Recorded as zustand
HD	Hail Damage	Maximum disassembly	BOSP, BAT, RT, FT, SSA, FAX, CHS, BSA	8	RAX (HD)	Standard
RD	Rear Damage	Rear axis repair	BOSP, RT, CORE, BSA	4	RAX (RD)	Unfallfahrzeug mit Heckschaden
TL	Total Loss	Material recovery	BOSP, BAT, RT, FT	4	CRE (TL)	Totalschaden
SA	Shock Absorber	Part recall	RT, FT, SSA, BSA	4	CSEB-NABS (SA)	Produktionsfehler an Stoßdaempfern

Cross-Scenario Data Analysis

This section presents the aggregated disassembly time statistics across all six scenarios. The analysis distinguishes between manual and automated tools.

Process Step Times

Tables 5 and 6 present the average handling and disassembly times per process step across all scenarios. Handling time (Table 5) includes both disassembly time (component removal) and movement time (transport to next buffer). It represents the total time from when a component enters a workstation until it exits. As outlined in Table 6, the disassembly time does not include the operator's moving time. It represents solely the duration during which the worker actively processes the component. Both raw and cleaned values are provided. Outliers were removed using the interquartile range (IQR) method to support the reproducibility of the results. Please note that due to the inconsistent quality of incoming products, there may be significant variations in disassembly times. In some cases, the disassembly of components proved challenging, often due to the presence of worn screws. This is an example of the unpredictable behavior that can occur in disassembly systems. Manual work was carried out by hand, while automated work was carried out using a cordless screwdriver.

Table 5. Average handling times per process step (in seconds)

Step	Operation	Tool	Statistics
			N	Mean	Std	Mean (clean)	Std (clean)	Outliers
S1	BOSP	Man	56	93	43	90	37	1
S2	BAT	Man	43	81	32	81	32	0
S3	RT	Man	37	118	63	105	33	3
		Auto	20	103	58	103	58	0
S4	FT	Man	31	126	60	120	50	1
		Auto	14	75	30	75	30	0
S5	FAXS	Man	18	197	62	197	62	0
		Auto	16	163	108	127	43	2
S6	SSA	Man	21	182	96	157	59	2
		Auto	18	121	25	121	25	0
S7	CHS/CORE	Man	20	159	57	159	57	0
		Auto	24	147	65	121	29	4
S8	BSA	Man	21	147	48	147	48	0
		Auto	22	125	51	125	51	0

Table 6. Average disassembly times per process step (in seconds)

Step	Operation	Tool	Statistics
			N	Mean	Std	Mean (clean)	Std (clean)	Outliers
S1	BOSP	Man	56	58	39	53	29	2
S2	BAT	Man	43	44	25	42	23	1
S3	RT	Man	37	79	63	67	31	2
		Auto	20	59	46	59	46	0
S4	FT	Man	31	81	59	75	48	1
		Auto	14	36	19	36	19	0
S5	FAXS	Man	18	149	52	141	41	1
		Auto	16	97	45	97	45	0
S6	SSA	Man	21	142	67	131	46	1
		Auto	18	83	33	83	33	0
S7	CHS/CORE	Man	20	125	52	125	52	0
		Auto	24	96	56	78	31	3
S8	BSA	Man	21	106	48	99	38	1
		Auto	22	83	41	83	41	0

Notes for Tables 5 and 6:

• Handling time (Table 5) = Disassembly time + movement time at workstation
• Disassembly time (Table 6) = Active processing time only, excludes movement to buffer
• N: Total number of observations across all scenarios
• Mean/Std: Original values including all data points
• Mean (Clean)/Std (clean): Values after outlier removal using IQR method (Q1 - 1.5×IQR, Q3 + 1.5×IQR)
• Outliers: Number of observations removed during cleaning
• Tool types: Man = Manual tools, Auto = Cordless screwdrivers

For more details about the analysis, see disassembly_and_handling_times_all_scenarios.xlsx

Citation

If you use this repository for your research or industry projects, please cite the following article:

@article{JORDAN2025277,
title = {Demonstrator-based implementation of an infrastructure for event data acquisition in disassembly material flows},
journal = {Procedia CIRP},
volume = {134},
pages = {277-282},
year = {2025},
note = {58th CIRP Conference on Manufacturing Systems 2025},
issn = {2212-8271},
doi = {https://doi.org/10.1016/j.procir.2025.03.040},
url = {https://www.sciencedirect.com/science/article/pii/S2212827125004810},
author = {Patrick Jordan and Lasse Streibel and Niklas Lindholm and Wesal Maroof and Susanne Vernim and Laura Goebel and Michael F. Zaeh},
}

License

This repository and its contents are licensed under the MIT License.

Acknowledgements

This research was conducted at the Institute for Machine Tools and Industrial Management (iwb) at the Technical University of Munich. This research was funded by the Federal Ministry of Economic Affairs and Energy (BMWE) as part of the "Car2Car – Kreislauffahige, nachhaltige Fahrzeugverwertungskonzepte" project (19S22007H). The corresponding author wants to thank the more than 25 scientific colleagues from the iwb who participated in the disassembly runs in January 2025. Additionally, the author would like to express his gratitude to Patrick Schwartz for his assistance in the creation of the material flow diagrams.

References

Jordan et al. 2024

Jordan, P., Kroeger, S., Streibel, L., Vernim, S., Zaeh, M.F. (2024). Concept for a data-based approach to support decision-making in tactical tasks for planning disassembly systems. Procedia CIRP, 122, 288–293. https://doi.org/10.1016/j.procir.2024.01.042

Jordan et al. 2025

Jordan, P., Streibel, L., Lindholm, N., Maroof, W., Vernim, S., Goebel, L., Zaeh, M.F. (2025). Demonstrator-based implementation of an infrastructure for event data acquisition in disassembly material flows. Procedia CIRP, 134, 277–282. https://doi.org/10.1016/j.procir.2025.03.040

---

For questions, suggestions, or collaboration opportunities, please contact the corresponding author or visit our institute website.