Implementing Machine Learning With LinePulse

Share on linkedin
Share on twitter


  • Support the engineers’ decision making process with machine learning tools via LinePulse


  • Accessed on-premise, in the cloud or via hybrid solution
  • Formats include CSV, JSON, HDF5, and SQL Database


  • Improved throughput
  • Reduced scrap and rework 
  • Minimized warranty costs


Automotive OEMs and Tier 1s need to minimize the cost of quality by reducing scrap and rework rates as well as warranty claims. In order to identify the underlying causes of such issues, they collect data from their production lines which engineers then analyze to determine which signals are most relevant for a given problem.

The number of signals involved and the databases that they populate are enormous, consisting of hundreds of thousands of individual data points. Traditionally, data analysis implementations upgrade PLCs (Programmable Logic Controllers) on the manufacturing line via industrial PCs on workstations in order to implement complex sensor data transformations.

The Problem

The algorithms used in these analyses are built by skilled manufacturing engineers who are familiar with the line. This process is time- and labor-intensive and requires regular adjustments to compensate for changes on the line. 

Moreover, while the volume of manufacturing data has grown, the tools to handle it have not caught up. Engineers still use traditional techniques to manipulate data and make sense of it, relying on old solutions to solve new problems. Manufacturing plants are becoming data centers, requiring robust analytics to extract usable information. 

This is where Acerta provides value. Acerta’s LinePulse offers a scalable analytics platform that offloads the data engineering and data science workload so that engineers can act on insights hidden in the data they are already collecting.

Here’s how it works.

Solution Process

1. Gathering Data & Selecting The Right Model

When engaging with a new client, Acerta begins by gathering information about the client’s manufacturing and data collection systems. This includes manufacturing data from the assembly line as well as testing data from end-of-line (EOL) testing stations and other quality processes. This data can be single-value or time-series, labelled or unlabelled (i.e., including tags or metadata regarding which units passed or failed), but ideally includes as much traceability and part history as possible. 

Common data formats include CSV, JSON, HDF5, and SQL Database. Manufacturing data can be stored in a key-value store, data warehouse, or data lake + catalogue, and Acerta can access that data on-premise, in the cloud or through a hybrid solution using direct uploads or ingestion via API. 

LinePulse ingests data from disparate sources and normalizes it before passing it on to the relevant module, depending on the application. The platform then finds the best model fit out of thousands of possible machine learning models, tuning and deploying it either in the cloud or in a connected application in a docker container on an industrial PC on the line.

2. Analyzing Data with LinePulse Modules

The LinePulse dashboard allows engineers to understand their line’s performance using Acerta’s machine learning models, which learn and adapt to the customers’ use case. The current suite of LinePulse modules includes Smart Line Analytics, Advanced Anomaly Detection, and Intelligent Component Selection.

Smart Line Analytics

Smart line analytics takes the normalized data and processes it using a set of machine learning models that explain the function of the line in order to determine which operations are most significant for a given failure mode. This accelerates root cause analysis for plant personnel by pruning the number of signals that need to be analyzed so that engineers can quickly identify what needs to be fixed. 

Advanced Anomaly Detection 

Deploying the Advanced Anomaly Detection module enables automakers to improve their end-of-line testing by flagging units exhibiting abnormal behaviour. Machine learning models define abnormality via training data consisting of normal (i.e., non-defective) units. As a result, this module is able to detect failing units even when they fall within normal limits.

Intelligent Component Selection 

Utilizing the data from the Smart Line Analytics or Advanced Anomaly Detection modules, LinePulse can generate automated recommendations during the assembly process via the Intelligent Component Selection module. By locating the causes of defects during assembly, LinePulse is able to identify key decision points on the line where future failures can be avoided. 

3. Monitoring Signals & Detecting Drift

As experienced line engineers know, data drifts as the manufacturing floor changes. Normally, engineers only notice this drift when it results in a problem on the line, forcing them to respond reactively. LinePulse uses signal monitoring, drift detection and smart customizable notifications to alert line engineers about drift before it becomes an issue. Moreover, LinePulse can respond proactively by triggering an automated model retraining process based on the latest sensor data.


Acerta’s LinePulse machine learning platform empowers manufacturing data to improve throughput and reduce scrap and rework rates. By improving end-of-line testing with machine learning, Acerta’s solution can reduce the number of warranty escapes from a production line, minimizing a manufacturer’s overall warranty costs.

One client used the Advanced Anomaly Detection module to reduce annual warranty costs by approximately €2M per plant. Another employed the complete LinePulse platform to reduce scrap and rework by 43% and increase line throughput by 25%. In another case, after using Smart Line Analytics on existing manufacturing lines to identify the root cause of failures, a client decided to implement LinePulse on a new line to guarantee a successful launch. 

How Can We Help?

This website uses cookies.