Seite wählen

Leider ist der Eintrag nur auf Amerikanisches Englisch verfügbar. Der Inhalt wird unten in einer verfügbaren Sprache angezeigt. Klicken Sie auf den Link, um die aktuelle Sprache zu ändern.

Leider ist der Eintrag nur auf Amerikanisches Englisch verfügbar. Der Inhalt wird unten in einer verfügbaren Sprache angezeigt. Klicken Sie auf den Link, um die aktuelle Sprache zu ändern.

As more products become smart and connected, software is emerging as the connective tissue for value creation, even for companies that sell physical goods. The convergence of the physical and digital worlds begins with sensors and sensory data, which automates and quantifies pattern tracking for both product distribution and customer behaviors in the physical world. Such data is becoming the currency of the Industrial Internet economy, and the foundation for new software-enabled services.

Ongoing improvements in sensor technologies – including miniaturization, connectivity options, performance, cost, and energy consumption – are making intelligent products more accessible.  As this level of digital infrastructure develops, companies will be able to take advantage of growing data streams to apply powerful analytics for insights that can enhance existing services, enrich customer experiences, offer new insights and create alternative revenue streams, not only through new products but also through entirely new business models.

Industrial IOT Based Asset Maintenance

Before we explore the topic of Automated Machine Learning Architectures, let’s consider the following use case of the Presenso solution for Predictive Maintenance:

A wind farm needs to predict whether a wind turbine generator will fail within the next few weeks.  The turbines are remotely located where it is expensive and time consuming for technicians to visit for inspections and repair. In the case of stalled operations, there is a high cost associated with downtime and revenue losses, replacing expensive parts and high labor costs.

The wind farm needed to build a machine learning-based predictive maintenance model.  Because of the costs of travel to remote locations, the model needed to be tuned to only pick up extremely likely failures.  The business requirement was more tolerant of a false negative relative to a false positive.  With only 1% of the data constituting failure, Presenso up-sampled the minority class and tuned the model parameters for high accuracy.

Do you want to learn more about this topic? See Failure Prediction under Big Data Constraints: How to Handle Imbalance Classes

Introduction to Presenso Autonomous IIOT Analytics

Machine Learning (ML) and Deep Learning (DL) are used in many fields, including computer vision, speech recognition, and machine translation. At the same time, novice engineers still struggle to effectively apply DL.  Multiple decisions need to be made:  the choice between dozens of available ML & DL algorithms preprocessing and cleaning methods, tuning the hyperparameters of the selected approaches and calibrating the number of trees or number of layers (and nodes at each layer) for the dataset.

How effectively ML and DL are applied makes the difference and has a high impact on performance. The application of machine learning is tedious and time-consuming and requires the expertise of big data engineers and data scientists (rare commodities in today’s labor market).

As more industrial plants need to use Machine Learning, there is a need for ML solutions that autonomously perform inference on a given dataset.

Do you want to learn more about this topic? See the Role of Automated Machine Learning in the Smart Factory by Waseem Ghrayeb

Presenso’s automated Machine and Deep Learning engine selects the optimal algorithm and hyperparameters in a data-driven way without any human intervention. In this article post, we describe how this is performed by the Presenso Auto- MDL system.

Models Selection, Hyperparameters Optimization, and Features Extraction

A well-known method for optimizing Machine Learning hyperparameters is the Bayesian optimization, which iterates the following steps:

  1. Build a probabilistic model to capture the relationship between hyperparameter settings and their performance
  2. Use the model to select optimal hyperparameter settings.
  3. Run the selected machine Learning model with those optimized hyperparameter settings.

This process can be generalized (to avoid overfitting) to jointly select algorithms, data preprocessing and data cleaning methods, and their hyperparameters as follows: the choices of classifier/predictor and preprocessing methods are top-level, categorical hyperparameters, and based on their settings the hyperparameters of the selected methods become active. The combined space can then be searched with Bayesian optimization methods that handle such high-dimensional, conditional spaces.

This Auto- MDL approach of using Bayesian optimization is used to automatically customize the optimal big data processing and unsupervised machine learning models to the appropriate industrial IoT analytics task.

Do you want to learn more about this topic? See the Five Steps to Implementing Unsupervised Big Data Machine Learning by Nir Dromi

Presenso Auto- MDL

Auto- MDL is Presenso’s implementation of the above idea. It contains a full Machine and Deep Learning pipeline which is responsible for missing values, categorical features, sparse and dense data, and rescaling/retreading /normalization of the data.

Next, the pipeline applies a preprocessing and cleaning algorithms and an ML/DL algorithm.

Auto- MDL includes dozens of ML algorithms, tens of preprocessing methods, and all their respective hyperparameters, yielding a total of hundreds of hyperparameters.

The optimizing performance in Auto- MDL space of hundreds of hyperparameters can be slow.  Presenso jumpstarts this process by using meta-learning to start the job from good hyperparameter settings originating from previous similar datasets (using a similarity function). When a new dataset is added, the algorithm looks for similar datasets as a starting point and applies the settings from the previous data set to the new one.

A second improvement was to automatically construct ensembles: instead of returning a single hyperparameter setting (as standard Bayesian optimization would), we automatically construct ensembles from the models trained during the Bayesian optimization. Specifically, we use model ensemble and stacking to create small, powerful ensembles with increased predictive power and robustness.


As more industrial plants migrate to the Smart Factory there is clearly a huge need for predictive asset maintenance.

The dearth of talented professionals with domain expertise in Deep Learning and Machine Learning will not be solved by increasing the supply of engineers and scientists.  A technology solution is required so that facilities can adopt and scale predictive industrial IOT throughout a facility.  Presenso Auto- MDL is an off-the-shelf solution for industrial plants that are transitioning to Industry 4.0 using a similar level of resources and capabilities when operating in the older environment.