Methods

The SPOTSeven Process Model

SpotSeven defines a process model, which enables systematic optimization of complex real-world problems. It can be seen as an extension and adaptation of the well-known SixSigma procedure. Although steps of the SpotSeven process are clearly specified, they will not be used in an algorithmic manner. SpotSeven leaves room for individual procedures. The seven steps can be characterized as follows:

1) Define

spot_1Customer and business requirements are specified from a high level point of view. The current situation, i.e., states and processes, is described in detail. Considering quality, time, and costs, the necessary steps are defined. Critical parameters and statistics are identified. The project team with clearly specified responsibilities is build.

As a result from the definition step, a tentative project charter with the following elements is available:

  • Project description
  • Project goals
  • Benefits (value)
  • Market analysis
  • Project scope
  • Responsibilities
  • Start and end date

The project charter can be developed during a workshop, in which decision makers from the company participate. The following steps are planed with the SpotSeven team. In general, each of the following steps is initialized with a one-day team workshop.

2) Data Acquisition and Measurement
spot_2The second phase is devoted to the selection of suitable methods, instruments, and processes to collect data from the key process variables, which were defined during the first step. Data, which are already available, are collected, prepared, and integrated at the same time. This step comprehends also a systematic gathering of measurement units, frequencies, and accuracies. Important input-/output relations are defined. Based on the current value of the output variable the aimed value is specified.

3) Modeling and Analysis
spot_3Functional relationships between in-/output parameters are determined. Causes and effects are prioritized. SpotSeven uses a hierarchical approach, i.e., models with different complexity are used in parallel. High-quality open source software such as R, as well as commercial products such as JMP or Data Modeler, are used during the whole process. We distinguish process- and data analysis. Process analysis visualizes qualitative relationships, e.g., by integrating cause-effect diagrams.  Data analysis is based on mathematical and statistical methods and relies on quantifiable measurement values. Methods from classical design of experiments are used to formulate statistical hypotheses about assumed functional relationships, e.g., y = f(x_1,x_2..) + e.  Statistical tests are applied to analyze these hypotheses. In addition to this classical approach from statistics, which define the baseline of our modeling and analysis, we use interactive visualizations, which represent an extended version of exploratory data analysis. With respect to the aimed value of the output variable, the current value is analyzed.

4) Optimization

spot_4

Sensitivity analysis is applied in order to determine suitable parameters for the optimization process.  Based on this parameter set, which might include categorical, ordinal or numerical parameter values, an optimization technique has to be chosen. A broad variety of optimization methods is available. Besides the well-known classical approaches from mathematical optimization, SpotSeven team members have a strong background in computational intelligence. Evolutionary algorithms, namely evolution strategies and genetic programming, are valuable complements to standard approaches. Moreover, SpotSeven includes methods for multi-objective optimization, because many real-world problems have to handle several, conflicting objectives, e.g., minimization of production costs while maximizing product quality.

Our goal is to generate an understandable, simple solution. Interpretability is an important factor in the SpotSeven process. In many real-world settings, simple solutions are more robust than complex, high level solutions. Furthermore, simple solutions extrapolate better and protect against overfitting. We have developed several tools which allow an individual selection between exactness and robustness of the solution.  After the model-based solution was determined and tested on the model hierarchy, a suitable solution is selected for implementation. A corresponding plan is developed and communicated with the practitioners.

5) Integration and Deployment
spot_5The optimal process parameters, which were determined during optimization on the model, are implemented in the real-world system. The process improvement is verified on the real system. If the real system does not allow these tests, model-based verification procedures will be used.

Operating and performance figures are defined in order to guarantee sustainability. The new settings are discussed with project leaders, technical experts and users to reach a high acceptance rate. If the new setting does not meet expectations, further analyses (step 3) might be useful.

6) Control
spot_6A plan, which describes actions in time to keep the system under control, is developed. Methods from statistical quality control are tools of choice [Mont08a].  The documented solutions are delivered to the customer. Workshops and training courses can be organized to discuss results and to train practitioners.

 

 

7) Meta Evaluation
spot_7SpotSeven itself contains a continuous improvement process. We ask for feedback from the customers and keep the contact alive, even after the project is finished. It is our passion to deliver the best results and we are interested to learn from the process.