a member of groupelephant.com
Machine Learning (ML) and Automation
SAPCONET’s Machine Learning & Automation services sees us partner with clients to develop intelligent, data driven solutions that maximise business efficiencies.
- We assist our clients in formulating and implementing AI strategies.
- We work closely with clients to unlock value in data through the innovative use of modern machine learning techniques.
- We help clients identify opportunities to optimise business processes though automation.
The application of AI to these data in a business sense are virtually limitless ranging from speech recognition and sentiment analysis, objection detection, anomaly detection, optical character recognition, optimisation, demand planning, predictive maintenance, early warning systems, to pattern recognition.
Common techniques in ML include
Natural Language Processing (NLP)
Natural Language Processing (NLP) is used for the analysis and synthesis of natural language and speech. Examples include speech recognition and sentiment analysis. Chat bots commonly use NLP techniques.
Optimisation
Optimisation is applicable in many areas using techniques to refine the output of a function or process. This is commonly used in time-series scenarios such as demand planning and forecasting.
Recommender Systems
Recommender systems use data to connect users to content, experiences, or information they seek in a more efficient and effective way.
Classification
Classification includes anomaly detection, object recognition, and segmentation and is typically used across many forms of media – from audio and video, to structured and unstructured data.
Need a bespoke AI solution? We’d love to partner with you, so please get in touch.
Recent SAPCONET ML developments
Optimisation of industrial production process
Smooth operation of conveyor belts which transport logs from loading point to chipper in the pulp industry affects the output of the process.
The challenge was to determine if computer vision techniques could be applied to detect problems at the loading point and control the industrial equipment accordingly.
The solution makes use of feature analysis to identify, and measure felled logs using multiple video feeds. The data is then used to control the speed of the load decks to optimise avoid jams.
The result was improved feed-rate in the wood chipping process and reduced manual workload.
Improved object detection accuracy in security context
The security industry in general experiences high volumes of false alarms inherent in visual object detection technology though the ubiquitous use of cameras.
The challenge was to reduce the number of false alarms in a multi-camera security operation within a gated community.
The solution uses the false alarms in a continuous improvement lifecycle using feedback loops to retrain custom object detection models.
The result substantially reduced the number of false alarms.
Retail store demand forecasting
Forecasting sales represents the holy grail to some degree in retail. The ability to predict sales accurately affects supply chain operations profoundly.
The challenge was to build a sales forecasting model on item and store data while making provision for promotions and campaigns with a thirty to ninety day forecast window.
The solution makes use of a time-series forecasting model trained on several years of sales, promotional, environmental, and economic historical data with a continual improvement pipeline for periodic adjustments.
The result is promising thus far with the concept models performing remarkably well.
Multi-model Edge processing
Consistent debarking of logs in the pulp industry influences costs directly.
The challenge was to replace a subjective manual process with automated computer vision techniques.
The solution involved continuous, real-time tracking and processing of visual data through video feeds using parallel custom machine learning models deployed on edge devices have allowed accurate measurement of the debarking process.
The result was a successful proof of concept demonstrating an edge deployment of the automated solution which allows accurate debarking measurements to be applied in optimising the debarking process.
Image processing for traffic control
License plate recognition is common in traffic and security contexts worldwide with many cloud and edge solution embedded in system and platforms.
The challenge was to use the recognised license plates to determine transit duration and to route traffic to destination points in an industrial wood-yard.
The solution is an implementation of Automatic License Plate Recognition that combined with a custom transit algorithm integrated to the control room.
The result allows the control office to direct vehicles to the correct location within the plant.