Category Archives: Machine Learning

LLM based voice Interaction in WinCC OA!

Posted on by

At our USA Tour, someone asked about voice interaction with WinCC OA – and luckily, I had exactly that showcase in my demo.

I recorded a quick demo (not perfectly, sorry), but it still shows how easy it is to bring LLMs into WinCC OA: capturing voice directly in the browser, sending it to an LLM via simple HTTP REST calls, and letting the system respond or act.

WinCC OA is open and flexible – that’s why use cases like this can be built so quickly.

And excuse the not-so-nice UI – I built it myself 😄

🎥 Check out the short demo!

🚀 MCP Server for WinCC Unified – Powered by GraphQL and GenAI!

Posted on by

Just gave myself a little nightly challenge: I implemented an MCP Server for WinCC Unified, based on its GraphQL Server.

https://github.com/vogler75/winccua-mcp-server

Thanks to the GraphQL server’s built-in documentation and its clearly defined data structures, it was surprisingly straightforward to generate most of the MCP server code — with the help from Gemini! 🙌

Super excited about how well this combination works — the power of Unified, GraphQL, and GenAI all together! 💡

The prompt for the example in the picture was: “logon with username1 and password1 and then fetch the values of the Meter Input logging tag of the last 10 minutes and plot it.”

Next time I asked Claude to forecast my solar PV production to help me decide whether it’s a good time to run my dryer.

It’s fascinating how you can simply ask questions like “Should I run my dryer now?” and get intelligent – really? 🧐 – responses based on actual and historical production data.

What I like about that:
✅ Natural conversation with my process data
✅ Real-time insights from WinCC Unified data
✅ AI-powered recommendations …

👀 But can I trust it? No, you do not know how it came to this forecast…

Dockerfile for Python 3.9 with OpenCV, MediaPipe, TensorFlow Lite and Coral Edge TPU

Posted on by

Dockerfile

FROM python:3.9-slim
RUN apt-get update && apt -y install curl gnupg libgl1-mesa-glx libglib2.0-0 && rm -rf /var/lib/apt/lists/*
RUN echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | tee /etc/apt/sources.list.d/coral-edgetpu.list 
RUN curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add - && apt-get update && apt-get install -y python3-tflite-runtime && rm -rf /var/lib/apt/lists/*
WORKDIR /app
COPY requirements.txt /
RUN pip install -r /requirements.txt
COPY * /app/

Requirements.txt

opencv-python
mediapipe

Python and WinCC OA…

Posted on by

Connected Python to WinCC OA through a Websocket Manager. Python programs can connect to WinCC OA and read/write datapoints. Communication is JSON based, it’s simple to use in Python, see examples below (ws://rocworks.no-ip.org can be used for tests, but will not be available all the time).

https://github.com/vogler75/oa4j-wss

  1. dpGet
  2. dpSet
  3. dpConnect
  4. dpQueryConnect
  5. dpGetPeriod
  6. … more functions will be implemented

Required Python modules:

  • pip3 install websocket-client
  • pip3 install matplotlib

############################################################
# Open Connection
############################################################
import json
import ssl
from websocket import create_connection
url='ws://rocworks.no-ip.org/winccoa?username=demo&password=demo'
ws = create_connection(url, sslopt={"cert_reqs": ssl.CERT_NONE})

############################################################
# dpGetPeriod
############################################################
cmd={'DpGetPeriod': {
 'Dps':['ExampleDP_Trend1.'],
 'T1': '2018-02-07T18:10:00.000', 
 'T2': '2018-02-07T23:59:59.999',
 'Count': 0, # Optional (Default=0)
 'Ts': 0 # Optional (0...no ts in result, 1...ts as ms since epoch, 2...ts as ISO8601)
 }}
ws.send(json.dumps(cmd))
res=json.loads(ws.recv())
#print(res)
if "System1:ExampleDP_Trend1.:_offline.._value" in res["DpGetPeriodResult"]["Values"]:
 values=res["DpGetPeriodResult"]["Values"]["System1:ExampleDP_Trend1.:_offline.._value"]
 print(values)
else:
 print("no data found")

# Plot result of dpGetPeriod
%matplotlib inline 
import matplotlib.pyplot as plt
plt.plot(values)
plt.ylabel('ExampleDP_Trend1.')
plt.show()

############################################################
# dpGet
############################################################
cmd={'DpGet': {'Dps':['ExampleDP_Trend1.', 'ExampleDP_Trend2.']}}
ws.send(json.dumps(cmd))
res=json.loads(ws.recv())
print(json.dumps(res, indent=4, sort_keys=True))

############################################################
# dpSet
############################################################
from random import randint
cmd={'DpSet': {'Wait': True, 
 'Values':[{'Dp':'ExampleDP_Trend1.','Value': randint(0, 9)}, 
 {'Dp':'ExampleDP_Trend2.','Value': randint(0, 9)}]}}
ws.send(json.dumps(cmd))
res=json.loads(ws.recv())
print(json.dumps(res, indent=4, sort_keys=True))

############################################################
# dpConnect
############################################################
from threading import Thread

def read():
    while True:
        res=json.loads(ws.recv())
        print(res)
Thread(target=read).start()
    
cmd={"DpConnect": {"Id": 1, "Dps": ["ExampleDP_Trend1."]}}
ws.send(json.dumps(cmd))

Deep Learning Neural Network with WinCC OA and Clojure…

Posted on by

To learn how deep learning works I decided to implement a Multilayer Neural Network with Backpropagation by my own in Clojure (I did NOT use a library like Tensor Flow or Deeplearning4j). The  link between Clojure and the SCADA system WinCC OA is oa4j. With that connection the Neural Network can be used and trained with sensor data collected by the SCADA system …