Introduction

MANUFACIA is an AI development tool for anomaly detection and predictive maintenance, based on Cross Compass’s expertise in using AI to solve problems in the manufacturing industry. It supports Image data, Time-series data and Vibration data from the manufacturing industry.

No special knowledge of AI is required to use MANUFACIA. Anyone can easily generate AI by simply following the on-screen instructions. The entire process, from AI generation to accuracy improvement, can be completed on the production floor.

MANUFACIA and Greenia are the trademark or a registered trademark in Japan and in other countries of Cross Compass Ltd. All other product names mentioned in the document are the trademark or a registered trademark of each mentioned company. Names of software, system, or products do not necessarily have trademark notice (®, TM).

MANUFACIA Users Manual

• 1. Before starting MANUFACIA
  • 1.1. Lexicon
  • 1.2. What MANUFACIA can do
    • 1.2.1. Supported data formats
      • 1.2.1.1. Time-series data, vibration data
      • 1.2.1.2. Image data
    • 1.2.2. AI model generated by MANUFACIA
      • 1.2.2.1. AI model types
      • 1.2.2.2. AI model selection
    • 1.2.3. Turn the generated AI models to account
  • 1.3. How to evaluate AI judgement
  • 1.4. Start MANUFACIA
    • 1.4.1. Recommended hardware specification
    • 1.4.2. Display language
• 2. Let’s start MANUFACIA 1
  • 2.1. MANUFACIA workflow
  • 2.2. Select an account
  • 2.3. Create a lab
  • 2.4. Register assets
    • 2.4.1. Select data files
      • 2.4.1.1. Create new from folder
      • 2.4.1.2. Create new from file
    • 2.4.2. Set tags
      • 2.4.2.1. How to use tags
        • 2.4.2.1.1. Tags automatically assigned
      • 2.4.2.2. Use of tags
      • 2.4.2.3. How to set tags
      • 2.4.2.4. Examples of tag settings
    • 2.4.3. Add files to the asset
• 3. Let’s start MANUFACIA 2
  • 3.1. Create a new project
  • 3.2. Select AI model type
  • 3.3. Select AI model training type
  • 3.4. Select tags for assets to use
  • 3.5. Label assets
    • 3.5.1. Label examples
  • 3.6. Split data
    • 3.6.1. Split data by asset tags
    • 3.6.2. Split data by percentage
  • 3.7. Define pre-processing (Time-series/vibration data)
    • 3.7.1. Read CSV
    • 3.7.2. Clamp (Optional)
    • 3.7.3. Sliding window (Vibration data only)
    • 3.7.4. Window function (Vibration data only)
    • 3.7.5. Normalize/Standardize
    • 3.7.6. FFT (Vibration data only)
    • 3.7.7. Interpolate (Optional)
    • 3.7.8. Binarize (Optional)
    • 3.7.9. Batch
  • 3.8. Define pre-processing (Image data)
    • 3.8.1. Load image
    • 3.8.2. Gray scale/Channel selection
    • 3.8.3. Crop (Optional)
    • 3.8.4. Normalize/Standardize
    • 3.8.5. Binarize (Optional)
    • 3.8.6. Batch
  • 3.9. Train models
• 4. Let’s start MANUFACIA 3
  • 4.1. Evaluate project
    • 4.1.1. Select AI model 1
    • 4.1.2. Select AI model 2
    • 4.1.3. Evaluate AI model
      • 4.1.3.1. Performance index of neural network
      • 4.1.3.2. Confusion matrix
      • 4.1.3.3. Histogram
      • 4.1.3.4. Validation sample list
      • 4.1.3.5. Training loss curve
      • 4.1.3.6. Generate reports
  • 4.2. Apply AI models
    • 4.2.1. Deploy AI models
    • 4.2.2. Test AI models
    • 4.2.3. Integrate AI models in a system
    • 4.2.4. Retrain AI models
      • 4.2.4.1. Add/remove models to/from favorite
      • 4.2.4.2. Add assets to the lab
      • 4.2.4.3. Create a project
      • 4.2.4.4. Add assets
• 5. FAQ (MANUFACIA Operation)
  • 5.1. Error occurs to upload data
  • 5.2. Training fails
  • 5.3. Adding assets takes time
  • 5.4. Low AI model accuracy
  • 5.5. No idea which AI model is good
  • 5.6. Cannot pause the training
  • 5.7. Recreate AI model by adding new data
  • 5.8. Want to change the tag assigned to assets
  • 5.9. Want to delete lab
  • 5.10. Want to delete projects
  • 5.11. Want to delete assets
  • 5.12. Models will not be plotted in the scatter diagram
  • 5.13. Labels cannot be added

Install manual

• 1. Deliverable
• 2. Recommended system requirements
• 3. Host machine port to use
• 4. Install
  • 4.1. Check the distribution
  • 4.2. Unzip and load images
  • 4.3. Confirmation of registered docker images
  • 4.4. Setup
  • 4.5. Confirmation of the IP address of the host machine
  • 4.6. Installation
• 5. Confirmation of application’s running
• 6. Uninstall
• 7. Reference materials (Driver/library)
  • 7.1. How to check versions
  • 7.2. How to uninstall
  • 7.3. How to install
    • 7.3.1. Docker related
    • 7.3.2. Install CUDA and Nvidia-Driver
      • 7.3.2.1. Updating MANUFACIA from v2.0/v2.1/v2.1.2
      • 7.3.2.2. Find available drivers for the server with GPU board
      • 7.3.2.3. Install Nvidia-Driver
      • 7.3.2.4. Install CUDA 11.1

Trouble Shooting

• What shall I do if I get all results with N/A after training?
  • Check if tag setting to upload assets/ labeling to assets were correctly done
  • For time-series/vibration datasets: Check if CSV file format is supported
  • Check if memory ran out when MANUFACIA starts
  • Check if other applications are using GPU
  • Check if the access to MANUFACIA server is via proxy server
  • Check Nvidia driver version
• What shall I do if MANUFACIA will not start by manufacia-ctrl start command?
  • Check if the MANUFACIA executable path is correctly set
  • Check if host setting is correct
  • Check if there are error messages in log files
  • Check driver/library versions are supported
• Assets cannot be partly/fully uploaded. What shall I do？
• Local disk is full. Which files may I delete?
  • Deleting the cache files
  • Removing the application’s log files
• How can I check if application containers are running?
• How to start MANUFACIA after a server reboot?
• How to get and restore backups?
  • Get a backup
  • Restore

FAQ (General)

• About MANUFACIA
• Purchasing and user support
• Recommended hardware specification
• Data preparation
• Misc.

Release notes

• Changes in v2.2.0
  • Introduced Matthews correlation coefficient for model performance index.
  • Confusion matrix.
  • Training loss curve, ROC curve and AUC.
  • Model detail report function.
  • ZIP file upload for assets.
  • Improved threshold / distance calculation for binary classification.
  • Histogram improvement
  • Histogram for supervised learning
  • Added a button to reset threshold / distance to the default setting
  • Speed up SmoothGrad calculation.
  • Greenia SDK limited for inference and acquiring results.
  • Updated Nvidia-Docker2/docker-compose versions.
  • Renamed text “test” to “validation”.
  • MANUFACIA version number.
  • Bug fix: Training with vibration datasets fails in unsupervised learning.
  • Bug fix: Inappropriate handling of models with threshold/distance of NaN or Inf
  • Bug fix: Wrong sample detail was displayed
  • Bug fix: Wrong percentage was displayed in labeling assets UI
  • Bug fix: Favorite model flag handling.
  • Bug fix: Missing on-mouse binding behavior in the confusion matrix. (multi-class classification)
  • Bug fix: Percentage of assets displayed for each label in label assets UI.
• Changes in v2.1.3
  • Updated CUDA to version 11
• Changes in v2.1.2
  • Bug fix: Project training result will not be displayed.
• Changes in v2.1.1
  • Accelerated training.
  • Improved SmoothGrad display.
  • Improved behavior to upload assets.
  • Changed initial axis parameter setting in model evaluation graph.
  • Changed axis scale for network size.
  • Added model creation time for each in model evaluation table.
  • Changed the preprocessing sequence of vibration datasets.
  • Changed preprocessing default setting.
  • MANUFACIA icon.
  • Bug fix: Although install-manufacia command fails, it will be displayed in console that it was successful.
  • Bug fix: Browsing models with “<>” keys in the model result page from unsupervised learning, histogram will be switched to confusion matrix.
• Changes in v2.1
  • ResNet50 and MobileNetV2 models were implemented for MANUFACIA 2D image recognition.
  • License authentication has been changed.
  • Network size was reduced.
  • Added support 8-bit depth image file.
  • CUDNN flag was enabled to accelerate training.
  • Bug Fix: manufacia-ctrl update will not always work.
  • Bug Fix: “manufacia-ctrl start” command creates an empty file called “1” every time it is run.

Image data augmentation tool

• 1. Introduction
  • 1.1. What is MANUFACIA Image Data Augmentation Tool?
  • 1.2. Operating environment
  • 1.3. Installation and how to start
    • 1.3.1. Windows
    • 1.3.2. Linux (Ubuntu)
  • 1.4. Open Source license information
• 2. Instructions
  • 2.1. Collect input image files in a directory
  • 2.2. Specifying input/output directory
  • 2.3. Set processing parameters
    • 2.3.1. Preprocessing parameters
    • 2.3.2. Augmentation parameters
    • 2.3.3. Postprocessing parameter
  • 2.4. Preview
  • 2.5. Start processing images
• 3. Change history
  • 3.1. ver 1.0.0

CSV format checker

• 1. Introduction
  • 1.1. What is CSV formar checker?
  • 1.2. Operating environment
  • 1.3. Installation and how to start
    • 1.3.1. Windows
  • 1.4. Deliverables
    • 1.4.1. Initial setting file
  • 1.5. Limitations
• 2. Instructions
  • 2.1. List of features
  • 2.2. Feature detail
    • 2.2.1. Show version information of the tool
    • 2.2.2. Format check
    • 2.2.3. Correct format
    • 2.2.4. Process data
      • 2.2.4.1. Divide file
      • 2.2.4.2. Extract rows
      • 2.2.4.3. Extract columns
      • 2.2.4.4. Thin out data
      • 2.2.4.5. Swap rows and columns
      • 2.2.4.6. Replace header line
    • 2.2.5. Additional options
      • 2.2.5.1. Specify input data path
      • 2.2.5.2. Rebuild time data
      • 2.2.5.3. Force data processing
• 3. Change history
  • 3.1. ver 0.14

AI model tester

• 1. Introduction
  • 1.1. What is AI model tester?

Contact / Disclaimer

• Contact us
• Disclaimer
  • Materials
  • Guarantees and responsibilities
  • Management of materials