Our team brings extensive know-how for various data sources available in a factory. They can be Historians, Energy Meters, Lab Quality Systems, MES, ERP, PLCs, SpreadSheets, Logs, SQL, PCs, Batch Reports, MTConnect, OPC-UA. Our Data Extraction Consultancy is categorized mainly into five parts.
The data which is collected from different sources may have dirty data, which is why the cleaning of data should be done before the data is loaded. The problem with polluted data is that there is no fixed way of dealing with it. The polluted values affect our performance and predictive capacity. Errors in the data have the potential to change all our statistical parameters. The way they interact with outliers once again affect our statistics. Conclusions can thus be misleading.
Many times, work order or product quality results are being captured manually, whereas automated systems are in place for sensors data, so combining the data creates logs of bad data
For handling bad data quality and faulty data, we leverage our powerful tool kit. It includes pre-written data-cleanup algorithmic modules such as sanity handling, missing handling, multicollinearity analysis, mahalanobis distance, data distribution check, infer best bucket etc. Once the data have been cleaned, it will produce precise results when the ML/DL algorithms are applied. Hence consistent data is essential for reliable decision making. We at Tvarit sanitise the data as surgically as possible to obtain the best possible solution.
The wave of Digitization and Data collection during the past years has forced every single company to focus on Data Collection. The biggest pain point of manufacturing companies as of today is to figure out which data is most fruitful. Further big data is being produced from Machinery as well, as thousands of sensors in your plant collect the data at the rate of every 1 second, sometimes even 1 millisecond. Therefore, valueable insight rather lie in “The Fruitful Data”, not in “Big Data”.
Intelligent Transformations such as FFT, Wavelet, Approximate Entropy etc can be applied on high-frequency data. For example, you are capturing Vibration data from a CNC Machine Spindle at the rate of 2KHZ which translates to a couple of GBs within a day. Applying “slot aggregation” becomes much easier as you can easily see that ~99% of times your CNC Machine Spindle is behaving normally and this “normal” data can be safely aggregated to the higher bucket (say 1 data point every 1 min), assuming no information loss. Now, the rest of the ~1% of the time, your CNC Machine Spindle is capturing Anomalies (during worn-out conditions or tool breaking conditions) which should not be aggregated at all, as that is “the Fruitful Data” and dropping the same will lead to information loss. This will allow this Data Compression from a sveral GBs to MBs of data without compromising accuracy.
Tvarit Experts have prior experience in process engineering plants where the calculation of precise set points of various parameters is very important to avoid any future anomalies. Our data scientists have built a ML/DL assisted recommendation engine to achieve that. Further, the Confidence Levels of these AI predicted setpoints are given while recommending users (shop floor engineers) with action items. Limits of input tweakable parameters are taken into consideration while creating these recommendations. Hence domain knowledge is incorporated into the ML/DL model and provision to users with sensible action items is ensured.