# About MANUFACIA
Q. What is MANUFACIA？
>A. MANUFACIA is an AI model generator which is primarily focused on manufacturing floor to offer anomaly detection or preventive maintenance of facilities. Generated AI model can be integrated into multiple types of devices or equipment.
Q. Which data is supported by MANUFACIA?
>A. MANUFACIA supports time-series, vibration or image data which are acquired in manufacturing floor.
Q. What are specific features of MANUFACIA?
>A. MANUFACIA features:
>- MANUFACIA can generate AI models in a simple workflow even without deep understanding of AI.
>- MANUFACIA has NN（Neural Network）implemented in it and users do not need to define features of datasets not as same as machine learning.
>- Generated AI models by MANUFACIA can be integrated into production equipment and be used for daily operation.
Q. I have very little idea about AI, but can I create AI models with MANUFACIA?
>A. To create AI models with MANUFACIA, no AI knowledge is necessary. With the help of intuitive UI, AI models can be easily generated without necessity of coding.
# Purchasing and user support
Q. How does the step or plan look like until purchasing?
>A. We offer several steps that you can evaluate MANUFACIA with your own data to be sure to decide to use MANUFACIA for your solution.
>- AI Factory: Free MANUFACIA result reporting service by using real data
A report will be delivered in about a week after receiving your data.
>- Trial: Evaluation license for about one to three months.
>- Purchase MANUFACIA
Q. How about user support?
>A. We offer full maintenance support and software updates after purchasing MANUFACIA with annual maintenance contract.
Q. How much will it cost for each plan, and how can I pay?
>A. Please contact us from [here](../contact.html#contact-us).
Q. It seems difficult to buy a server and purchase MANUFACIA. What shall I do?
>A. We also offer free-of-charge plans. With these free plans, you can evaluate MANUFACIA without cost and then decide to purchase.
Furthermore, you can use a usual laptop PC to create AI models and evaluate them, if you will use MANUFACIA-CPU for your initial evaluation.
(recommended-hardware-specification)=
# Recommended hardware specification
Q. Training/prediction environment: Host(server) PC system requirements?
>A. For the training, server needs to process many mathematical operations. Here are the system requirements for server machines.
>GPU version
>- CPU: Intel® Core™ i7-9700 (8 core, 8 threads, 4.9GHz) or equivalent
>- Memory: 32GB (16GB x 2)
>- Storage: 4TB SSD or more (2TB storage is also fine depending on data amount)
>- Network: Gigabit
>- Graphic card: NVIDIA® GeForce RTX™ 2080 Ti 11GB-GDDR6
(GeForce 1070 is fine only for time-series/vibration data)
>- OS：Ubuntu 18.04 LTS & CUDA 11.1
>- Virtual environment: Docker (ver 19.03)
>CPU version
>- CPU: Intel® Core™ i5-5257U (2 core, 3.1GHz) or equivalent, CPU with AVX/AVX2/FMA instruction set support
>- Memory: 8GB
>- Storage: 256GB (SSD recommended) or more
>- OS：Ubuntu 18.04 LTS (Including WSL2 on Windows10 Home/Pro)
>- Virtual Environment: Docker (19.03)
```{note}
It was confirmed that MANUFACIA will work on AWS, but it was not yet tested on other cloud services.
```
Q. MANUFACIA operating environment: Client PC system requirements?
>A. On the client PC, it only needs to start a browser. Usual office PCs are good enough. On following browsers MANUFACIA operation was confirmed.
>- Google Chrome 94.0.4606.61 (64bit)
>- Microsoft Edge 94.0.992.31 (64bit)
Q. Production environment: Requirements for edge device/machine?
>A. Depending on what to do, it may need GPU for some cases and only CPU for other cases.
# Data preparation
Q. Which data formats are supported for training?
>A. Following data formats are supported.
>- Time-series/vibration data: CSV format
>- Image data: RGB or gray scale. PNG/BMP/ (JPG: not recommended for accuracy)
Q. Is it necessary to align data length for CSV data?
>A. Yes. MANUFACIA could principally handle data with different length but for relatively small data. It is recommended to align data length in advance.
Q. Which kind of image data do we need?
>A. The bigger the image data for training, the longer the training will take. Therefore, **it is recommended to use 224x224 pixel size and at maximum 512x512 (the upper limit in preprocessing)**. If you use much bigger image data without resizing, it will just take much longer to upload image files.
Moreover, it is also recommended to prepare **image data in which anomaly part, such as contamination, scratch/blemish, porosity, discoloration or dirt is in the size of 5 to 10% of the width/height of the image (several 10 pixels)**, so that it can be can be identified by human eyes.
Q. How many normal/anomaly data do we need to prepare?
>A. Depending image data, MANUFACIA can find feature value from datasets with several 10 files. However, **it is recommended to prepare 50 to 300 files for each of normal/anomaly**.
Q. We do not have much anomaly data, what can we do?
>A. In the production flow, there are many cases that good amount of anomaly data is missing. In MANUFACIA it is also possible to create AI models only from normal data in unsupervised learning. However, even in that case, anomaly data for validation is necessary.
Q. How about the security?
>A. MANUFACIA can be installed in your environment and you can create AI models without sending out your important data to outside.
Q. What to do if only a little amount of data is available?
>A. For customers that use image data, we also offer image data augmentation tool with MANUFACIA. This tool can increase the number of data.
# Misc.
Q. What about issues that cannot be solved by MANUFACIA?
>A. In Cross Compass we also develop customized AI. Moreover, we offer consulting, education or seminars.