Important diagnostic calls lost on desensitised crew
Macsea’s Dexter machinery health monitoring system screen shot
Wendy Laursen explains how analysis of fault calls reveals how crew become desensitised to equipment alarms and how electrical equipment in particular is emerging as a problem.
US-based Macsea, developer of the Dexter machinery health monitoring software system, was recently asked by a customer to perform a forensic investigation of diagnostic data retrieved from a vessel that contained a large number of faults. In the data sample provided, there were 291 diagnostic fault calls generated and 56% of these resulted from failed or erratic sensors. Another 32% were due to improper operation of a standby lube oil pump. A specific diagnostic had been included in Dexter’s knowledgebase to detect when the standby pump was being operated in non-manoeuvring situations when it shouldn’t be. Pump operation during these times meant it would be consuming additional energy and creating excessive pressure in the lube oil system, increasing the risk of seal failures and oil leaks.
“Clearly the crew was ignoring this energy conservation and maintenance avoidance directive,” says Kevin Logan, president of Macsea. “The customer was shocked to learn of these results. Although they had been involved in CBM-type systems for some time, they had never seen data like this before and had no formal remediation procedures in place to address either the sensor or the energy conservation problems.”
Typically, Dexter is integrated into a ship’s machinery control system and operates on Windows-based workstations located throughout the machinery spaces. The software acquires real-time machinery data directly from the control system and its diagnostic and prognostic software modules perform real-time equipment health analysis and reporting to the crew.
Mr Logan believes that ship-specific grooming is needed to fine-tune condition monitoring systems for maximum diagnostic robustness. Excessive alarm generation will desensitise operators to legitimate fault calls. Top-down acceptance of the CBM concept and proper training is therefore important. “The lesson learned in this case study is that successful CBM implementation begins with the ‘eyes and ears’ of the machinery plant. Ignoring sensor issues will only result in failed projects and none of the cost-saving benefits that CBM can provide. The continuation of alarm flooding situations indicates to us that a customer doesn’t really get it – you can’t diagnose or predict equipment failures from bad sensor data. It sounds simple, but we see sensor issues all the time on ships.”
Beyond sensors, diagnostic robustness is directly related to the artificial intelligence inferencing techniques used by the CBM system. Early expert systems employed rule-based or logic-based reasoning. Although relatively easy to build, these systems are vulnerable to bad input data such as results from sensor problems. Artificial neural network (ANN) technology addresses this problem because it is tolerant of noisy or incomplete input patterns. ANN models learn from training examples that relate machinery faults to their abnormal symptom patterns. An exact match between what the ANN has learned through its training and what it senses in real-time from the machinery plant is not necessary in order to diagnose a fault. The pattern-matching feature isolates the closest matches from the learned relationships and assigns them a probability score based on the number of matching symptoms.
Dexter uses experiential and engineering knowledge of subject matter experts to construct a diagnostic data base for ANN learning. Comprehensive failure mode and effects analysis on critical equipment targeted for diagnostic coverage is undertaken. This gives more complete coverage in the shortest possible time, while focusing on mission-critical or historically unreliable equipment that consume excessive maintenance resources.
Dexter has been taken up by both commercial operators and the US Navy and has been operating on some vessels for over 15 years. The majority of applications involve main propulsion diesels, diesel gensets and gas turbine engines but also includes auxiliary monitoring of systems such as seawater and freshwater cooling, lube oil, and fuel oil systems.
Macsea is currently applying its proven diagnostic technology to shipboard electrical systems, particularly integrated power systems (IPS) with variable frequency drive (VFD). “We see a growing and urgent need to provide assistance to shipboard crews who are struggling to understand how these new complex systems work and keep them running safely and efficiently,” says Mr Logan. “Reliability problems are starting to surface with all-electric and hybrid-electric power systems that involve multiple diesel and gas turbine generators feeding into shipboard power grids that provide power to propulsion motors, dynamic positioning thrusters, weapon systems and essentially every piece of electrical equipment aboard.”
In recent years, there has been an alarming increase in shipboard electrical failures, including fires on ships employing electric propulsion, with VFD-related issues pertaining to harmonic filter systems, electrical noise, and variable speed AC motors. Excessive harmonic distortion can create severe overheating of various electrical equipment including components such as capacitors, bearings, laminations and winding insulation on generators, transformers, and induction motors, stators and rotors of fixed speed electric motors, cables, and filters for electromagnetic compatibility or transient suppression.
“Some serious issues are evolving with the increased use of IPS. We see a need to make these complex electrical systems smarter such that they can better monitor their own health and assist not only the crew, but perhaps even the control systems in detecting faults and taking remedial action to avoid catastrophic failures. Dexter’s neural network-based reasoning capabilities can be readily integrated into the IPS design and may enhance their ‘self-healing’ capabilities.”