Knowledge graph dataset of product structures and disassembly sequences of 10 EV battery packs

DATA SET CONTENTS

The data set contains product structure data for ten battery packs from electric vehicles (components, fixations, and their relationships to one another) as well as disassembly sequence data, which includes the order of the disassembly steps and the specific disassembly operations required for each step. The data set includes packs from the Fiat 500e, Chevrolet Spark, Hyundai IONIQ 5, BMW 530e, Chevrolet Volt, Volvo EX30, BMW 330e, MG ZS, Mitsubishi Outlander, and BMW i3.
In addition, for each component, the dataset records which components are visible at the time of its disassembly. Together with the included sampling function, this allows the output of an object recognition model to be simulated. This enables the application of disassembly planning algorithms without first having to implement an object recognition model, which would otherwise be required.
The dataset consists of four CSV files, an importer file in JSON format, and a Python script. The CSV files contain the raw data and form the core of the dataset; the other two files are auxiliary files that support the usability of the data.
• battery_catalogue.csv: Overview file containing all batteries and additional details on model, year of manufacture, and video sources
• parts.csv: All recorded components, including class, assignment to a battery, and visibility list
• fixations.csv: All fasteners, along with the components they connect and the direction of the connection
• operations.csv: All disassembly operations, including assignment to a component or fastener and their sequence
• neo4j_importer_model.json: Graph schema and mapping of CSV columns for the Neo4j Data Importer
• sample_confidences.py: Sampling function for detection confidences

DATA COLLECTION

The data was collected manually. The battery packs from the Volvo EX30, BMW 330e, and Mitsubishi Outlander were disassembled manually, the remaining packs were analyzed using publicly available disassembly videos, which are listed in battery_catalogue.csv.
The terminology for battery components varies in the literature and by manufacturer. For this dataset, the naming of components was therefore standardized, so the terms used may differ from those in the videos or other sources.

STRUCTURE AND CONVENTIONS

The battery ID serves as the prefix for all associated IDs. All components, connectors, and operations with IDs in the 1xxx range belong to the battery with ID 1, all with IDs in the 2xxx range belong to the battery with ID 2, and so on. The files are also linked to one another via the following fields: battery_id in parts.csv and fixations.csv refers to the id in battery_catalogue.csv; fromPartId and toPartId in fixations.csv, as well as part_id in operations.csv, refer to the id in parts.csv; fixation_id in operations.csv refers to the id in fixations.csv; and next_op_id refers to the id of the next operation.
When reusing the data, several conventions should be noted. The visible_parts field contains the visible components, including the component itself; the individual entries are separated by a plus sign. For operations, exactly one of the two fields, fixation_id or part_id, is populated, depending on whether the operation refers to a fastener or a component. The next_op_id field maps a linear path for each battery; alternative, equivalent disassembly sequences are not included. The disassembly operations used are lift_off, unscrew, unplug, unclip, and cut. However, connections of the fixed class do not have their own operation, as they are removed together with the parent assembly. Assemblies such as modules or electronic units are each treated as a single component and are not further disassembled.

DATA USAGE

The CSV files are plain text files and can be imported using a text editor, a spreadsheet program, or programmatically. To use the data as a knowledge graph, the file neo4j_importer_model.json can be loaded into the Neo4j Data Importer. However, the underlying representation is not tied to a specific graph database. Using the relationships described above, the four CSV files can also be loaded into a general-purpose graph library, or into another graph database, and processed using common graph algorithms.
The script sample_confidences.py assigns detection confidences to the visible components and their associated connection elements, which are drawn from class-specific, left-truncated logit-normal mixed distributions. Python 3 is required.
The dataset is particularly suitable for developing and evaluating algorithms for disassembly sequence generation that do not require CAD data, for comparing the product structures and disassembly sequences of different battery packs, and for investigating the relationship between components, fastening technology, and disassembly effort as a basis for disassembly-friendly design. The formalization used, consisting of components, fasteners, operations, and sequences, is not battery-specific and can also be applied to other products.