Manufacturing Defect Dataset

This is a ready-made synthetic dataset for machine learning and quality control analysis. It contains 51,267 rows across 3 tables, detailing product information, defect occurrences, and associated image metadata. Preview rows are available for free.

  • last updated 28 Sept 2026
  • by GoMask
  • Dataset includes 5000 product units and 27,235 defect instances.
  • Defect data spans from 2023-01-01 to 2024-12-31.
  • Features include product material, manufacturing line, defect type, severity, and detection method.
  • Image data includes URL, format, resolution, lighting condition, and file size.
  • Data is synthetic, generated by GoMask DataFactory.

At a glance

  • 3tables
  • 51,267rows
  • 17columns
  • Jan 2023 – Dec 2024date range

The 3 tables

preview and data dictionary per table

Products products · table · 5,000 rows

Information about manufactured product units.

Preview

First 10 of 5,000 rows of the Products table
product_idintegerproduct_namestringmaterial_typestringmanufacturing_linestring
801Liam Heavy Steel BeamSteelLine A
802Olivia Structural PlateSteelLine A
803Noah Steel GirderSteelLine A
804Emma Steel SupportSteelLine A
805Oliver Steel FrameSteelLine A
806Ava Steel ChannelSteelLine A
807Elijah Steel BarSteelLine A
808Sophia Steel RodSteelLine A
809James Steel BracketSteelLine A
810Isabella Steel CoilSteelLine A
10 of 5,000 rows · 4 columns

Data dictionary

Data dictionary for the Products table
columntypedescriptionexamplenull %
product_idintegerUnique identifier for each manufactured product unit.unique8010%
product_namestringDescriptive SKU or model name for the manufactured part.Liam Heavy Steel Beam0%
material_typestringPrimary base material composition used to manufacture the product.Steel0%
manufacturing_linestringDesignated factory line or workstation where unit was fabricated.Line A0%

Defects defects · fact table · 27,235 rows

Details of each recorded defect instance.

Preview

First 10 of 27,235 rows of the Defects table
defect_idintegerdefect_typestringseveritystringdetection_methodstringtimestampdatetimeproduct_idinteger
1ScratchMinorAutomated Scan2023-08-30 17:55:101553
2OtherMinorSensor Reading2024-06-12 17:11:341969
3Cosmetic BlemishMinorAutomated Scan2023-07-06 08:49:451770
4OtherMinorAutomated Scan2024-06-28 09:48:132593
5Cosmetic BlemishMinorVisual Inspection2023-12-21 17:57:081469
6OtherMinorSensor Reading2023-03-21 14:49:112037
7OtherMinorAutomated Scan2024-08-16 08:55:12871
8OtherMinorSensor Reading2023-05-11 20:00:361938
9OtherMinorSensor Reading2024-07-25 12:58:263279
10Cosmetic BlemishMinorAutomated Scan2024-01-09 16:50:252070
10 of 27,235 rows · 6 columns

Data dictionary

Data dictionary for the Defects table
columntypedescriptionexamplenull %
defect_idintegerPrimary key identifier for each recorded defect instance.10%
defect_typestringSpecific classification of the observed manufacturing defect.Scratch0%
severitystringSeverity assessment level indicating the impact on product usability and standards.Minor0%
detection_methodstringQuality assurance inspection method used to discover the defect.Automated Scan0%
timestampdatetimeTimestamp indicating when the defect was identified during inspection.2023-08-30 17:55:100%
product_idintegerForeign key to products.product_id.15530%

Defect images defect_images · fact table · 19,032 rows

Metadata for images associated with defect instances.

Preview

First 2 of 19,032 rows of the Defect images table
image_idintegerimage_urlstringimage_formatstringresolution_mpdecimallighting_conditionstringfile_size_kbintegerdefect_idinteger
175https://qc-storage.factory-mesh.internal/images/defect_3/175.pngPNG21.27Darkfield Ring45903
307https://qc-storage.factory-mesh.internal/images/defect_10/307.pngPNG13.03Darkfield Ring369210
2 of 19,032 rows · 7 columns

Data dictionary

Data dictionary for the Defect images table
columntypedescriptionexamplenull %
image_idintegerUnique identifier for each defect inspection image record.10%
image_urlstringHosted web storage URL locating the defect inspection image artifact.https://qc-storage.factory-mesh.internal/images/defect_17191/1.png0%
image_formatstringImage file container format.PNG0%
resolution_mpdecimalResolution of the inspection camera sensor in megapixels.27.20%
lighting_conditionstringOptical lighting configuration used to highlight the defect during capture.Darkfield Ring0%
file_size_kbintegerCompressed size of the visual artifact in kilobytes.78110%
defect_idintegerForeign key to defects.defect_id.171910%

How the tables join

  • defects.product_id references products.product_idmany to one: each Defects row points to one Products row
  • defect_images.defect_id references defects.defect_idmany to one: each Defect images row points to one Defects row

Questions to answer with it

  1. Can a machine learning model predict defect severity based on product features?

    Consider joining products and defects tables and analyzing the relationship between material_type, manufacturing_line and severity.

    tables: products, defects

  2. What is the distribution of defect types across different manufacturing lines?

    Join products and defects, then group by manufacturing_line and defect_type to count occurrences.

    tables: products, defects

  3. How does the frequency of defects change over time for specific product batches?

    Join products and defects, filter by a specific product_id or product_name, and group by date (extracted from timestamp) to count defects.

    tables: products, defects

  4. Identify the top 3 most common defect types and their average severity.

    Group defects by defect_type, count occurrences, and analyze the distribution of severity for each type.

    tables: defects

Starter SQL

run against this data before publishing

Table names match the SQLite file and the SQL script.

Top 5 Most Common Defect Types

sql
SELECT defect_type, COUNT(*) as defect_count FROM defects GROUP BY defect_type ORDER BY defect_count DESC LIMIT 5;

Defects by Manufacturing Line and Severity

sql
SELECT p.manufacturing_line, d.severity, COUNT(d.defect_id) as defect_count FROM defects d JOIN products p ON d.product_id = p.product_id GROUP BY p.manufacturing_line, d.severity ORDER BY p.manufacturing_line, defect_count DESC LIMIT 20;

Average Image Resolution per Defect Type

sql
SELECT d.defect_type, AVG(di.resolution_mp) as avg_resolution FROM defects d JOIN defect_images di ON d.defect_id = di.defect_id GROUP BY d.defect_type ORDER BY avg_resolution DESC LIMIT 20;

Defect Count Over Time (Monthly)

sql
SELECT STRFTIME('%Y-%m', timestamp) as month, COUNT(defect_id) as defect_count FROM defects GROUP BY month ORDER BY month LIMIT 20;

Defects with Critical Severity and their Product Material

sql
SELECT p.material_type, d.defect_type, d.severity FROM defects d JOIN products p ON d.product_id = p.product_id WHERE d.severity = 'Critical' LIMIT 20;

Load it with pandas

python
import pandas as pd

# Unzip the CSV download first: one file per table
products = pd.read_csv("products.csv")
defects = pd.read_csv("defects.csv")
defect_images = pd.read_csv("defect_images.csv")

# Join defects to products
df = defects.merge(products, left_on="product_id", right_on="product_id", how="left", suffixes=("", "_products"))
print(df.groupby("material_type").size().sort_values(ascending=False))

Using it in your tool

Excel
Load each table into a separate Excel sheet. Use Power Query to join 'defects' and 'products' tables on product_id. Create a PivotTable from the joined data to analyze defect counts by manufacturing line and material type.
Power BI
Load all tables. Create relationships: defects.product_id to products.product_id, and defect_images.defect_id to defects.defect_id. Create measures for Total Defects (COUNT(defects[defect_id])) and Average Severity Score (e.g., mapping Minor=1, Moderate=2, Critical=3).
SQL
Load the SQLite or SQL dump into your preferred SQL environment. The primary join paths are defects.product_id -> products.product_id and defect_images.defect_id -> defects.defect_id. Use foreign keys to link related data.
Python
Use pandas to load CSV or SQLite data. Join tables using pd.merge(). Analyze defect distributions using value_counts() and group by operations. For ML, use libraries like scikit-learn to build predictive models.

Formats available

  • CSV (zip)One CSV file per table, zipped
  • Excel workbookOne worksheet per table
  • SQLite databaseA ready-to-query database file with every table
  • Parquet (zip)One Parquet file per table, zipped
  • SQL scriptCREATE TABLE with primary and foreign keys, then INSERTs
  • CSVA single CSV file
  • JSONA single JSON file

How this data was generated

Synthetic data. Every row was generated: no real people, customers or companies are in this dataset.

Synthetic data generated by GoMask DataFactory from a relational blueprint: keys, links and rules are enforced in code, text columns are filled by a language model. No real people, companies or transactions.

  • The dataset was generated using GoMask DataFactory.
  • Synthetic data allows for controlled generation of defect types and severities.
  • Relationships between product attributes and defect characteristics are simulated.
  • Checked by an automated quality gate: unique keys, no orphan foreign keys, required columns filled, declared rules and date ranges (realism score 100).

Limitations

  • The data is synthetic and may not perfectly represent all real-world defect complexities.
  • Image data is represented by URLs and metadata, not the actual image files.
  • The dataset does not include information on corrective actions taken for defects.
  • Distributions and correlations are modelled, not measured from real records.

15 SQL practice questions for the Manufacturing Defect Dataset

Written for this data and checked against it: 12 SQL and 3 Power BI exercises from beginner to advanced. Try the first three free; the practice kit has all 15, a starter .sql file and an answer key with every expected result.

  1. 01BeginnerFreeSELECT · WHERE · ORDER BY · LIMIT

    Critical defects since July 2024

    Return every defect with severity 'Critical' detected on or after 2024-07-01. Show the columns defect_id, defect_type and timestamp, ordered by timestamp ascending and then defect_id ascending. Return only the first 100 rows.

    Tables: defects · your result: 3 columns, 100 rows
  2. 02BeginnerFreeGROUP BY · COUNT · ORDER BY

    Defects per defect type

    Count the defects for each defect_type. Return the columns defect_type and defect_count, ordered by defect_count descending and then defect_type ascending.

    Tables: defects · your result: 2 columns, 7 rows
  3. 03BeginnerFreeGROUP BY · COUNT · multi-column grouping

    Products per material and line

    Count the products for each combination of material_type and manufacturing_line. Return the columns material_type, manufacturing_line and product_count, ordered by material_type and then manufacturing_line.

    Tables: products · your result: 3 columns, 4 rows
  1. 04Average image size by formatBeginner
  2. 05Defects by material bar chartPower BI
  3. 06Critical rate by manufacturing lineIntermediate
  4. 07Monthly defects with image coverageIntermediate
  5. 08Line and material pairs with many cracksIntermediate
  6. 09Critical defects without imagesIntermediate
  7. 10Average resolution by lighting matrixPower BI
  8. 11Top two defect types per lineAdvanced
  9. 12Month-over-month defect changeAdvanced
  10. 13Running total of image storageAdvanced
  11. 14Percent of total by detection methodAdvanced
  12. 15Percent of total defects by severityPower BI

Get the practice kit: the full dataset, all 15 exercises, a starter .sql and the answer key. One-time $3.99.

Get the practice kit · $3.99

blueprint · manufacturing-defect-dataset

Scale this dataset

Same tables. As many rows as you need.

Open the blueprint behind these 3 tables in Data Factory: keep the relationships, change a column, and generate it at the size you need.

  • 200,000 rows
  • 1,000,000 rows
Scale this dataset in Data Factory
Tables
products, defects, defect_images
Licence
yours to use, including commercially
API slug
manufacturing-defect-dataset

What should your data show?

Preview 20 rows free
No signup. No card.