Data Scientists | Intelligent Data Infrastructure

Our Intelligent Data Infrastructure Addresses Costly Issues with Industrial Production Lines

Currently, there is a lot of hype about AI in industrial production.

Yet, before you think about deploying an AI driven solution, you should first ascertain if you have got an appropriate stack of data.

Our experience, which dates back to 2018, tells us: in most cases this is not the case. As a consequence, AI solutions won’t help cutting cost. To the contrary, they will produce additional cost. PANDA | IDI is a unique kit of software (PANDA | Drift), Hardware (Edge | Device)
and optional algorithms.

PANDA | IDI lets you Operate with Data in line with both, AI and your Particular Needs

Your machines and sensors should produce meaningful multi-modal data in a homogenous way. Often, you will require a vast amount of high-frequency data to arrive at meaningful conclusions. Yet, your plant might stop working altogether, if those data would jam your factory or machine data networks.

PANDA’s Intelligent Two-Pronged Edge and Network - Based Data Solution Addresses this very Challenge:

It avoids data jams paralyzing your factory while at the same time allowing aggregated data getting stored and algorithms getting deployed on your edge.

Standard Software

Microservices

The Data Flow between Services is Organized by Streams

PANDA | DRIFT uses a MQTT Broker to connect different services and data flows with an efficient and flexible pub-sub system.

How does PANDA | DRIFT Work?

Each service is configured with a JSON file and a docker compose file connects the services. Add custom code with templates.

Choose a microservices and configure it with a JSON files.

BUILD

Write a docker compose file to route the services.

RUN

Build & deploy a docker image to a device.

Write a JSON file to configure a service


version: '3'
services:

  mqtt-broker:
    image: registry.panda.technology/drift/devops/mqtt_broker_amd64
    ports:
      - "1883:1883"

  trigger-service:
    image: registry.panda.technology/drift/system_services/trigger_service:dev
    command: python3 -m trigger_service
    environment:
      MQTT_BROKER_URL: tcp://mqtt-broker:1883
      CONFIGURATION_PATH: /var/config/trigger-service.json
      LOG_LEVEL: INFO
    volumes:
      - ./config:/var/config
    depends_on:
      - mqtt-broker

  hello-service:
    build: .
    ports:
      - "8000:8000"


version: '3.8'
services:
  mqtt-broker:
    image: registry.panda.technology/drift/devops/mqtt_broker_amd64
    ports:
      - "1883:1883"

  trigger-service:
    image: registry.panda.technology/drift/system_services/trigger_service:dev
    command: python3 -m trigger_service
    environment:
      CONFIGURATION_PATH: /var/config/trigger_service.json
      MQTT_BROKER_URL: tcp://mqtt-broker:1883
    depends_on:
      - mqtt-broker
    volumes:
      - ./config:/var/config/

  input-proxy-service:
    image: registry.panda.technology/drift/system_services/input_proxy_service:dev
    command: drift_input_proxy_service
    volumes:
      - ./config:/var/config
      - /dev:/dev
    ipc: host
    environment:
      MQTT_BROKER_URL: tcp://mqtt-broker:1883
      CONFIGURATION_PATH: /var/config/input_proxy_service.json
      MQTT_CLIENT_ID: drift_input_proxy_service
      OUTPUT_MQTT_TOPIC: camera/compressed
      INPUT_MQTT_TOPIC: camera_image_notify_1
      LOG_LEVEL: DEBUG
    depends_on:
      - mqtt-broker

  daheng:
    image: registry.panda.technology/drift/io_services/shmemory_camera_service:dev-daheng
    volumes:
      - ./config:/var/config
      - /dev:/dev
    ipc: host
    privileged: true
    environment:
      CONFIGURATION_PATH: /var/config/daheng.json
      MQTT_BROKER_URL: tcp://mqtt-broker:1883
      MQTT_CLIENT_ID: shared_memory_camera_service
      TRIGGER_INPUT_MQTT_TOPIC: trigger
      TRIGGER_OUTPUT_MQTT_TOPIC: camera_image_notify
      SHARED_MEMORY_BASE_NAME: shared_image
      SHARED_MUTEX_BASE_NAME: shared_mutex
    depends_on:
      - trigger-service

  data-exporer:
    image: registry.panda.technology/drift/data_explorer/data_explorer_backend:dev
    depends_on:
      - mqtt-broker
    environment:
      MQTT_BROKER_URL: tcp://mqtt-broker:1883
      MQTT_CLIENT_ID: data_explorer_backend-1
    ports:
      - "8000:8000"


version: '3'
services:

  mqtt-broker:
    image: registry.panda.technology/drift/devops/mqtt_broker_amd64
    ports:
      - "1883:1883"

  trigger-service:
    image: registry.panda.technology/drift/system_services/trigger_service:dev
    command: python3 -m trigger_service
    environment:
      MQTT_BROKER_URL: tcp://mqtt-broker:1883
      CONFIGURATION_PATH: /var/config/trigger-service.json
      LOG_LEVEL: INFO
    volumes:
      - ./config:/var/config
    depends_on:
      - mqtt-broker

  opcua-client:
    image: registry.panda.technology/drift/io_services/opc-ua-client-service:v1.1.2
    restart: always
    environment:
      MQTT_BROKER_URL: tcp://mqtt-broker:1883
      MQTT_CLIENT_ID: opcua-client
      OUTPUT_MQTT_TOPIC: opcua-data
      CONFIGURATION_PATH: /var/config/opcua.json
      LOG_LEVEL: INFO
    volumes:
      - config:/var/config/

  influx-writer:
    image: registry.panda.technology/drift/system_services/influx_writer:dev
    volumes:
      - ./config:/var/config
    environment:
      MQTT_BROKER_URL: tcp://mqtt-broker:1883
      MQTT_CLIENT_ID: influx-writer
      CONFIGURATION_PATH: /var/config/influx-writer.json
      INFLUX_URL: http://influxdb:8086
      INFLUX_ORG: ferrum
      INFLUX_TOKEN: ${INFLUX_TOKEN}
    depends_on:
      - mqtt-broker
      - influxdb


  influxdb:
    image: influxdb:2.0-alpine
    environment:
      DOCKER_INFLUXDB_INIT_MODE: setup
      DOCKER_INFLUXDB_INIT_BUCKET: data
    volumes:
      - "./influxdb:/var/lib/influxdb2"
    healthcheck:
      test: [ "CMD", "influx", "-version" ]
    ports:
      - "8086:8086"

check full documentation

Write a docker compose file to connect the services.


{
  "endpoint_url": "opc.tcp://127.0.0.1:4840",
  "subscription_parameters": {
    "publishing_interval": 500,
    "monitoring_sample_rate": 100
  },
  "buffer_size": 100,
  "authentication": {
    "type": "anonymous"
  },
  "security": {
    "mode": "none",
    "policy": "None"
  },
  "variables": [
    "ns=1;s=sinusoid_noise",
    "ns=1;s=sinusoid"
  ]
}


{
  "triggering": {
    "capture_rate_ms": 1000
   },
   "triggers":[
      "SIMPLE_TRIGGER",
      "INTERVAL_TRIGGER"
   ]
}


{
  "type": "DAHENG",
  "capture_timeout_ms": 2000,
  "camera": {
    "settings": [
      {
        "enable": true,
        "instances": 2,
        "uid": "NE0200070087",
        "trigger_source": "software_trigger",
        "exposure_auto": "continuous",
        "gain_auto": "continuous",
        "balance_ratio_auto": "continuous"
      }
    ]
  }
}

PANDA | DRIFT Bundles a Best Selection of Open Source Software and ML Libraries.

MinIO

MinIO offers high-performance, S3 compatible object storage.

Native to Kubernetes, MinIO is the only object storage suite available on every public cloud, every Kubernetes distribution, the private cloud and the edge.

MinIO is software-defined and is 100% open source under GNU AGPL v3.

Visit Homepage

InfluxDb

InfluxDB is an open-source (MIT licensed) time series database.

It provides simple HTTP API to write data as time points and aggregate them as time series by Flux/SQL-like query languages.

Visit Homepage

OpenCV

OpenCV is an open source (BSD licensed) library for computer vision applications.

It provides a common data structure for images and classical and machine learning algorithms to process images.

Visit Homepage

Python

Python is a simple, universal and concise programming language, which is used in particular for creating scripts and the uniform addressing of various frameworks, which may themselves be written and compiled in other languages.

Visit Homepage

Mosquitto

The Eclipse Mosquitto project is an open source (EPL/EDL licensed) message broker that implements the MQTT protocol.

It provides a C library for implementing MQTT clients on all edge devices from low power single board computers to full servers.

Visit Homepage

PyTorch

PyTorch is an open source (modified BSD license) machine learning framework that provides fast tensor computing and deep neural network building blocks with an underlying automatic differentiation system.

Visit Homepage

Grafana

Grafana is an open source (under AGPL 3.0 license) web application for data visualization.

It can take data with all the popular databases and visualize it as graphs in the web browser. Also, it has alarms and can be used for sending notification in several ways.

Visit Homepage

TensorFlow

TensorFlow is an open source (under Apache-2.0 License) machine learning framework that provides fast tensor computing and deep neural network building blocks with an underlying automatic differentiation system.

Visit Homepage

FastAPI

FastAPI is an open source (MIT licensed) framework for building Web APIs with Python 3.6+ based on standard Python type hints.

Visit Homepage

Scikit-learn

Scikit-learn is an open source (BSD-3-Clause licenced) machine learning library.

In addition to the algorithms it also provides various tools for model fitting, data preprocessing, model selection, model evaluation, and many other utilities.

Visit Homepage

Panda|Metric

PANDA | METRIC is an open source (MPL 2.0 licensed) framework for dependency mining based on the concept of metric spaces, which makes it possible to use and combine and connect arbitrary data by topological principles.

Visit Homepage

ca. 4 entries

NGINX

NGINX is opens source software (BSD licenced), which is commonly used as a reverser proxy to hide the internal microservice infrastructure and provide a secure HTTP\WS public interface.

Visit Homepage

Kubernetes

Kubernetes is the most widely used open source (Apache 2 license) container orchestrating tool.

It allows to bundle few edge devices into one cluster and automate the deployment of micro-service applications.

Visit Homepage

Open62541

Open62541 is an open source (MPL 2.0 licensed) implementation of the OPC UA standard to build lightweight clients and servers for the OPC UA protocol.

Visit Homepage

ProtoBuf

Protocol buffers (BSD 3-Clause License) provide a language- and platform-neutral, serializing structured for data up to a few megabytes in size. It’s like JSON, except it's smaller and faster, and it generates native language bindings to transfer data over network and for long-term storage.

Visit Homepage

Continuous Integration and Deployment for DevOps

Keep overview and order with continuous integration and deployment. Monitor your decentralized solution with dashboarding tools.

Use PANDA | DRIFT for on-Premise Applications, no Cloud Needed

From data acquisition to fully automated evaluation and pattern recognition with the help of artificial intelligence: Build your own independed industrial AI platform embedded in the control cabinet.

Sensors

Embedded devices

Touch panels

Install PANDA | DRIFT on an Edge Device and Get Immediately Started with AI

PANDA | DRIFT offers a complete solution to handle and manage Linux based edge devices. Algorithms, Drivers and Databases, you get everything in a small setup on industrial hardware.

Interface Carrier

Camera & 4 sensor ports
14.0 interfaces (WLAN/LAN)
1TB SSD mass storage

Multimode Amplifyer Board

Amplification
Filtering
A-D Convesrion

Compute Board

8-Core CPU
8 TOPS TPU Co-Processor
32GB RAM

Start with PANDA's Preconfigured Reference Hardware

PANDA provides a portfolio of highly optimized hardware for DRIFT

USB Power Banks

Plug & Play

4 channel

48.000 Hz

A/D-Converter
ADS8866 Texas

16 Bit

The basis for the application of AI methods is excellent data quality. Industrial sensors of any kind can be processed directly on PANDA’s edge devices by using PANDAs sensor extensions.

Contact PANDA sales

USB Power Banks

Plug & Play

4 channel

48.000 Hz

A/D-Converter
ADS8866 Texas

16 Bit

The basis for the application of AI methods is excellent data quality. Any industrial sensors can be processed directly on edge devices by using PANDAs sensor extensions.

Contact PANDA sales

USB Power Banks

Plug & Play

4 channel

48.000 Hz

A/D-Converter
ADS8866 Texas

16 Bit

The basis for the application of AI methods is excellent data quality. Any indutrial sensors can be processe directly in edge devices by using PANDAs sensor extensions.

Contact PANDA

The Digital Transformation of your Production Starts with Data Quality

To ensure that all information about the machine condition and the product quality can be recorded easily, DRIFT offers streaming interfaces for sensors. Your equipment can be retrofit with sensors, without modifying the existing machine control system.

Use PANDA | DRIFT to Synchronize and Store Data from Different Sources

One of the most striking features of DRIFT is the ability to sync different data sources that contain high or low frequency information or even images.

Real Time

Async Time

Stream buffers

Real Time

Cameras

Machine Data

Synchronization

Storage

AI Ready Data Aquisition with PANDA | DRIFT

You Want to Deploy your own Algorithm?

PANDA | DRIFT is 100% compatible with Python and ML libraries

Use Preconfigured AI Services

PANDA | DRIFT provides starting points for common use cases

Computer Vision (simple application 1 - one day to make it work)

1 camera
1 DNN algorithm for object detection
1 Custom code to use bounding boxes to black out workers
Inline - visualization app

Agile AI Development Made Easy

Start developing with your favorite tools, then seamlessly switch to high efficient ML-implementations by utilizing DRIFT's streaming format.

apply algorithm on full resolution with standard tools

internal storing and streaming

apply optimized algorithm in high performance low latency mode directly on compressed data

Process Large Data Efficiently with PANDA | DRIFT’s Wavelet Technology

DRIFT’s internal streaming data is based on wavelet decompositions. It enables a universal treatment of matrix - like data formats by utilizing a Bfloat16 sparse matrix representation.

20x

Less storage space for data

10x

Speed-up for ML algorithms

WBS data format for time series

decompose

80% of values
are zeros

compose

90% of values = zeros

WBS data format for images

decompose

compose

DRIFT’s WBS data format compared to alternative ML formats

in RAM (floats, random access, direct access for algorithms and linear algebra operations)

[npy]

on SSD or Network-traffic
(binary, no direct access for algorithms)

.npy

.jpg

.wb

kBytes (less is better)

Learn More about Solutions that are Built with PANDA | DRIFT

Automated Quality inspection

Computer Vision building blocks

The perfect start for AI in production. A computer vision system is a camera-based quality control system via pattern recognition and without a cloud system. Industrial interfaces for automation include OPC-UA or Profinet.

Condition Monitoring

Anomaly detection and life time prediction

Empower the maintenance team with modern pattern recognition methods to indentify problems before they oocure. Use DRIFT to build your contition monitoring and predictive maintenance platform.

Dependency Mining

Learn how to use Metric Spaces to find dependencies in any data format

A mathematical framework developed specifically for applications in industrial production that is based on topolocial theory.

Documentation

Check the full documenation of DRIFT microservices

Learn how to build your data-driven AI platform using DRIFT microservices and connected open source software.